Shane Butler: Hey, everyone!
Sravya Madipalli: Hello, hello. Hello?
Shane Butler: Let’s… Start as we always do. If you can hear us, then drop where you’re from in the chat.
Sravya Madipalli: Yep.
Shane Butler: I’m based out of South Lake Tahoe. Who else we have here? What’s this… what’s the time zone spread we got going on? Nice. UK, Nairobi, Ohio, Seattle… You know, no one is from Tahoe in these things, ever. No one’s ever on these from Tahoe. I don’t know what everyone here is doing.
Sravya Madipalli: I get a bunch of Bay Area, for sure.
Shane Butler: Yeah. Yeah. UK… Nice. Where else we got going on? I know there’s more than, 6 of you.
Sravya Madipalli: Yep.
Shane Butler: Oh, I gotta admit some more people. That’s my problem. Nice. India… hey, everyone. We’re just dropping.
Sravya Madipalli: Aloha.
Shane Butler: Locations in the chat as we wake up the chat, and people, come in…
Sravya Madipalli: Let me check.
Shane Butler: We’re in the waiting room off, actually.
Sravya Madipalli: Yeah, so that everyone can come in.
Shane Butler: Nice. Hey, Attendee. You didn’t get, tired of us yesterday? Came for another round?
Sravya Madipalli: I see, actually, a couple of faces that, were there yesterday’s session, too, our Intro to Cloud Code.
Shane Butler: Yeah, it’s awesome. Cool. Well, I think we got, got about 25 people coming in. We still have people rolling in, but… Probably do a quick, intros and stuff, Savia. I think a lot of people here have come to our other ones, but you never know.
Sravya Madipalli: Yeah, sure. I could start a bit. Hello everyone, Stravio Maripali, I am, I have around 15 years of experience as data science. I started off at Microsoft, and later on moved to eBay and Nextdoor, and now, the most recent, tenure was at Superhuman. It was previously called Grammarly, they just changed the name to Superhuman. So that’s about me. So if you heard me talk through my experience, you saw me talk about Nextdoor. That was the last but one company I worked at. That’s where Shane Hai, who is our other co-founder of AI Analyst Lab, he’s not here today because he has another conflict, but…
Shane Butler: He has a job. He has a day job. What a nerd. He’s not… you know, he’s not gonna watch this. It’s cool. I always cut out the beginning of the ways. Don Kai, take the leap.
Sravya Madipalli: I love how we make… keep making fun of… Hi, hi.
Shane Butler: Whatever he’s not using.
Sravya Madipalli: not around. Oof! Yeah, so hi, Shane and myself, we three of us. Wanted to do something as soon as we decided to leave next door, to stay in touch. We loved working with each other, so we started a podcast. If you all don’t know, please, we called it Data Neighbor. We keep uploading some of our previous workshops to that YouTube as well. check out the channel. We had a bunch of interviews and podcasts we’ve done in that as well. Yeah, and then we moved on, to actually sharing all the knowledge that we acquired over the years. I would say, cumulatively, it’s around, like, 40 years of experience, all three of us. So, we decided to share all this knowledge with the world, starting with Maven.
So, and welcome. This is a free workshop, and we do have a lot of paid courses as well that we’ll get through later, but… I think… Yeah, Shane, go ahead with your intro, and then we can get started on the workshop.
Shane Butler: Cool, yeah, I just dropped that, I dropped our YouTube in there, too. Our YouTube has all the old workshops, like Stravia said, and it has, like, a couple years of our podcast. Though we haven’t done it for a few months. Yeah, I’m Shane, one of the co-founders of AI Analyst Lab. Yeah, I’ve been in data science for about 10 years. And, yeah, I’d say the last two years kind of focused… took a shift away from, like, pure product data science to more, AI evaluation and agentic analytics work. So, yeah, excited to have you all here today, and, several of you. For a second day in a row, also see a bunch of familiar names from our, From our full cohorts, so that’s cool.
I’ll… I’ll let you take the… take the reins of this, Ravia.
Sravya Madipalli: Awesome, okay, let me get started. So, feel free to keep, you know, posting your questions or anything while I run the session. Shane, like, you know, feel free to stop me in between. And we’ll get started. Okay, so we, this is more of a newer session, and probably we have a nice, cozy crowd today. We decided it’d be pretty cool to share about, all the different types of, you know, areas that we could just publish analysis. With Cloud Core, and hopefully we’ll have enough time to go through demos of multiple styles of, you know, publishing analysis with Cloud Core. Yeah. And we basically take one analysis, and we try to, share with you a Google Doc, a Notion page, and Google Slides Deck.
All of it coming from this analysis. Within minutes, and hopefully no reformatting, we never know with live sessions how things look like. But, you know, let’s try and get going. Okay. Have you seen this happen with you? You all see my screen, right? Do you see my.
Shane Butler: Yes, so it’s like, you finished the analysis, I think that?
Sravya Madipalli: Okay, awesome. Yeah. So, I’d love to know you all, like, raise your hand if this happened to you. You’ve finished the analysis, the thinking is done, you basically know what you want to present, the answer’s clear, but… you still have, oh, the 2 hours of work, I don’t know what I was thinking. I think it sometimes even takes up to a day, or even more, right? To get the analysis into the right format, into the right talk, into getting the right visualizations, and you know, all of that. Have this happen with you. You have the story, but… It takes a ton of time to put that together. Oh, this is, some fun messages, but yeah. Yeah, right? So, hoping, we’ll share some techniques with you.
This is something that Shane High and I have incredibly benefited from Claude Code, on, how we get the content into a deck format in, like. You know, minutes. But you need to have all the scaffolding ready, you need to have all the types of formatting that you’d like the deck to look like, and you know, all of that ready, obviously. That takes time to get the system set up, but once you know how you want it, what you want it, and you have some templates and frameworks on how you’d like the analysis to show up. It actually is way less amount of time. To get this done. Okay. Cool. So, let’s make this basically concrete, right?
So, think about your last analysis, the actual thinking, the exploration, the insight and recommendation, you know? Maybe all of that takes, like, 30 to 60 minutes for you to, like, you know, understand and go through, right? But now, the 30-60 minutes, by that, what I meant was to, like, once you have everything ready, to know what you want, how you want the story to look like, right? But… How you basically get the packaging done, the Google Doc with proper formatting, the charts look presentable, does the titles of the chart look well, where do you want to place the chart, you know, all of that stuff, right? That’s where you take sometimes even more than a simple analysis.
So, every person, like, in your team pays the same amount of tax every time. This is something that I would say that’s agnostic to human data, right? Even if different types of content, the packaging actually takes really long than the amount of thinking that you need to do. And the real problem is actually, you know, it not… it’s not the speed, it’s… I would probably call it, like, the sequence. The problem isn’t that people are, like, slow at formatting. It’s basically one manual step after the other, and all of these sequential steps basically takes a ton of time. And then, you basically are doing it by hand, right?
You… are… you… go through one after the other, ensure that, does the slide one, or, like, the first part of the dog, does it resonate well with the second part of the dog? Are things coherent with each other? And, you know, all of that. So there’s a lot of, like, back and forth that you’re doing. And this is what, you know, chain, like, takes the time. So, what are we trying to do here? We basically have the idea of analyzing it once and repackaging it for the audience, depending on the type of audience that you’re going to tackle.
And, ensuring that you have less manual step-to-step work, but all the styles of formatting, the templates of how you’d like the, you know, the doc to look like, or the analysis to look like, you have all of that pre-decided within frameworks already, so that once you have a final analysis, and once you know the story you’d like to tell, it’s literally a minute’s, worth of time for Claude to, you know, push something into, like, an analysis format. Okay, so… basically, every format has a different style of job. This is the demo, in this demo, that’s what we’re trying to do.
But obviously, at your work, for the type of work that you’re doing, for the type of stakeholders you work with, this is going to change for you. But in the demo, this is what we are going to tackle, right? So, this is, like, almost like an audience format map, where, basically, we are gonna have, like, a key insight. Each format has a different job. The Google Doc is going to be the doc that you share with your VP, where it’s most like a decision record. All it has is executive summary, what are the findings, what is, the recommendation. Like, make it as simple as possible, VPs don’t have enough time. All they need is, give me what you’re trying to tell me and why. Right?
So, that’s what we are trying to share with the Google Doc, today. It’s basically the type of, you know, analysis or output you’d like to share with the VP. Now, what about the Notion Doc? Notion Doc… In this demo, we’re going to share it like… it’s like a living wiki, wiki page. You have your analysis, you’ve done tons of work, right? You have multiple layers within the analysis, you have different charts you’d like to show, you in fact, might even want to show some of the code that was run. With, like, you know, you want to share your source doc or something with, like, a Databricks notebook or something like that. Or maybe you want to copy-paste the stuff.
I tried… to do one variety here today, but you could choose the type of variety you’d like to share with your team, right? So Notion is going to be way more detailed, and contains all different things that you would like to share with your teammate. So that they could, you know, help you review the analysis and ensure that you have something really solid, and there’s, like, nothing that, that’s not validated with your team, right? So that’s the Notion style of doc that we are going to share. And then, the third one is basically going to be the slides.
So this is basically, like, the presentation style, that imagine you get 5 minutes in your team meeting or in your, like, you know, all hands of your org or something. What would you share as part of that in those 5 minutes? So those will be the slides that you review, 5 or 10 minutes, basically, the slides that you review, one message per slide. Headlines that states the point, and charts that carry the proof. So these are the three different type of formats, the three different jobs, based on the audience that you’re going to, you know, tackle this messaging with, right? And based on that, you change each of them.
Which basically means it’s the same analysis, but different shape for each reader. Never, you know, the same text that passed 3 times. It basically has a different flavor based on the type of audience and the type of meeting, or if it’s async, or if it’s, like, an in-person meeting that you share. Okay. So, the four things, every published prompt names are, like, these are the four things, right? I don’t… I kind of explained it to you. I’ll try to be brief here as well. It’s basically, in your prompt, you want to share these things. You want to share the audience. Who is the audience? Is it the VP? Is it the team wiki? Is it a stakeholder review? And then the second one is the format.
What type of, you know, output are you looking for? I, today, in the demo, I’m going to share 3 types of output. outputs, like I already said, it’s going to be a Google Doc, there’s going to be a Notion page, and there’s going to be a slide deck. But guess what? Like Shane joked in the message today, there’s so many of these analysis that get shared in a Slack message, right? Or Teams, if you have Teams going on. You have, like, an insight, there’s a question, you share a visualization with, like, a message. So that’s another style of sharing analysis, or sharing insights, right? And also email, our typical, traditional way of sharing.
Today, I don’t do the Slack and email, but these are, like, multiple formats you could share your insights and analysis with, right? And another very important thing you want to share in your prompt, for publishing and Claude Code is the length and depth, right? You want to understand, the detail you want to share in your analysis. Like I said, if it’s going to be for a VP, you don’t, need to share all these details, right? You basically need to share the things that the VP cares about, probably a recommendation. And that’s it.
But if it’s your team, you want to share all the details that they could help review stuff, and you’re not sharing something that’s flawed, and maybe if it’s your teammate, broader, broader team with the engineers and product, you want to do something in between, right? And, yeah, that’s, follows into the fourth, framework piece here, what to include, right? So, exec summary, the data table information, the methodology of how you came about. coming up with these things, those are too detailed, right, for a VP. So, you basically ensure, and understand, what is the amount of detail you need, for what type of audience, for what format, and what length.
And all of these need to be part of Your publish prompt, so that it helps you, you know, get the right format for the right amount of people in the right place. Okay, so let’s get started. What I want to share with you, because we don’t have, like, hours and hours of time with each other, I decided what I’ll do is basically what we generally, you know, are read before we publish the analysis. Before we publish the analysis, we have all the charts that are needed that we think are interesting, right? This is something I’ve done, so, as a people manager for so many years. If you work, even if you mentor, like other data scientists, you see this a bunch, right?
Especially with junior data scientists is. When you have an analysis, you keep looking at hundreds of things, and you have so many visualizations, and you think most of them are interesting, some of them are really interesting, and when I worked with my data scientists. especially juniors, they just put all that stuff in the deck from the get-go. And that’s the hardest part, especially as a mentor or a people manager, is to get those 60 slide, 50-slide deck into a 15-slide deck.
Shane Butler: Sorry, yeah, I had, last year… I had… this guy’s very smart, and he’s a friend of mine, and he just… he only did this this one time, but he was at a… out of grad school, and starting… he did his first analysis with us, and I was like, he’s gonna turn on it. And his deck had 200 charts in it.
Sravya Madipalli: Oh, my God.
Shane Butler: 100 charts, and each of the charts were, like, cut, all these different, like, charts within charts. And, like, we got it down to 7 charts, like, that’s how much stuff was… unnecessary. there, but I got, like, what he was… he was going after. He was like, oh, if I’m in this meeting, and then they asked me about this, like, what if they think about this? I’m like, yeah, totally, but, like. Dude, no one’s gonna look at 200 charts. Wait, just have that on your computer, and you can pull one up.
Sravya Madipalli: You know, what I’ve seen is a lot of people, especially, even happens with me as well, even after all these years, right, is the hard part is to figure out what to pick. So… there are, like, 3 stages. You start off with the question, and you figure out the question, you have the answers. You need to decide, out of hundreds of insights, what is the insight you’d like to share? That’s the first decision you make. as a data person, or even a product person, what is something you’d like to share to start with? And then, what visualization you’d like to share that insight with, because you could have multiple ways of sharing the same thing, right?
So those are the tough and tricky decisions that you need to make at the tail end of your analysis, or tail end of your presentation, whatever it is. I see a question from Attendee. Attendee, do you want to ask this, in the chat? The question that you posed in the chat?
Attendee: So, the question is more like all this information, what you’re gathering and publishing or showcasing it to the leadership team, right? It’s still, in my opinion, it’s a one-way traffic where the information is gathered and being shared. Now that you mentioned about this tool, Google Doc, or Notion, or Slack, or whatever it is, what I’m trying to find out is, is there a way to take the feed from each of these data points back into your slide deck, or whatever tool that you’re using, so that, you know, there’s a refinement of information?
Sravya Madipalli: Absolutely, Attendee. Like, I mean, you’re basically talking about you have multiple narratives that you came up with in multiple areas, right? And can we leverage the other narrative, into, like, you know, when you probably are sharing with your VP, they would probably want to know this other angle that you already built for a different audience, right? And do we leverage each of them? Is that your question, Attendee?
Attendee: That is definitely one way of looking at it, and the question was more towards, you know, feed what we get from leadership, or for that matter, audience. How do I take it back into my columns. process.
Sravya Madipalli: Okay.
Shane Butler: Yeah, you can give us feedback directly where you’re kind of like. I… a lot of the ways that we iterate on stuff when we’re building is we’ll do, like, recorded sessions, and then we’ll bring that back in and be like, hey, what are all the things we need to update versus, then how would that reflect in code? We actually do that a lot with our courses. We have, you know, yesterday.
Attendee: lunch.
Shane Butler: Yesterday, we ran, like, the 3-hour workshop, and the first thing I did afterwards was I had Cloud Code analyze All of the chat and all of the, the transcript, and later I’ll have them do their reviews, too, in the Slack messages to see, like, hey, what’s all the stuff that we can improve for next time? And there’s also, like, There is also a personalization aspect to it when you’re… when you’re… delivering to, like, smaller audiences, where something I did at my work was I had, profiles for each of my primary stakeholders, where I actually analyzed the types of documents that they wrote, the types of Slack messages that they had.
Than how they, like, talked to me, and that way I had an idea, like, hey, if I am pris… if I’m talking to… Like this guy. John, I know that these are what he cares about.
Attendee: It’s more like a personality MD, if I may say, just like the way I have a clot.md, so I can think about if I’m presenting to a specific VP, so we have to sort of understand the tone of… and style of how they present, or what information they consume, and what… Yeah. Put it back into the personality MD. Okay.
Shane Butler: Exactly, because there’s, like, you know, we can think of in generalities of, like, hey, we’re speaking to a leadership audience, but, like. your leadership audience at your company could be totally different than a leadership audience at someone else… at another company, right? So, yeah. It’s a really good question, the feedback loop stuff is pretty powerful.
Sravya Madipalli: Absolutely. So, totally agree, Attendee, sorry I didn’t get it fully right.
Attendee: The first time.
Sravya Madipalli: So, this is like a V1, right? I’m trying to share with you, you have an opinion, you have a story together. How do you… use Claude Code, leverage Claude Code to Share all these insights that you have. into multiple styles of presentations, right? Multiple styles of analysis and presentation. So this is V1. What you’re talking about is the most important thing. Once you present, how does that get into the next version, right? So, I can give my example, and then we can move on into the demo, is that, what I try to do, all my… meetings have a note taker. We used otter.ai.
And I basically have MCP with Otter in my cloud code as well, and as soon as we are done with the presentation, I give the order notetaker back to Claude, and I give my perspective of how the meeting went, and give it the transcript of how the meeting went, and help me come up with the, changes that we’d like to do, and still the open-ended questions, and what the V2 should be about, and does it even make sense to have a V2 or not. This is literally something that I have in my flow as part of any presentation, analysis, like, any meeting, in fact. It doesn’t have to be in presentation, right? Anything you share, it’s going to be tackled with feedback. And how are we handling the feedback?
And some of the feedback is actually not live in a meeting. The feedback is actually in commence, right? So there’s so many of my presentations, of my team’s presentations, that go in… so, like, Notion, we have something called Coda, that’s what I used to use at my previous place. So, all the commence of leadership and all of that. I make Claude Code read through comments and come up with an idea of either answering within the commands as an answer, or coming up with another V2 version of the doc that answers those questions in, like, a summary. Yeah.
Attendee: Wow, that’s a great turnaround good people. Thank you so much.
Sravya Madipalli: No problem at all. Awesome. Okay, so, we are dealing with the V1, right? The V1 of the analysis here, and let’s say we have the analysis ready, we know what we want to present, we have tons of charts, like the 18 charts, the whole chart library. I try to make it 18, to be concise for this demo, but we all know that it could get up to, like, 200, that Shane said, that, we could, you know, have all of those. For us, at our, exposure, or at our hand, that we could pick, for the type of analysis we’d like to publish in Google Docs or Slides and others, right?
So, I also want to share with you how we came with the analysis as well, so that you understand a little bit about the system behind this analyst that we built. We can start with that in the demo. And then we’ll go with demoing different, you know, styles of analysis, or different, places in which we could share and publish the numbers, okay? Let’s get started. I’ll stop sharing my screen, and I’ll share my Cloud Code with you all. Give me a minute. Okay. By the way, if you have any questions, like Attendee, feel free to, you know, ask the questions. I just dropped a link.
Shane Butler: to that. open source version of the repo of what Schwabi is public, and then… Pull up here.
Sravya Madipalli: Yeah. Let me share my Start with. Okay. Okay, what you’re looking at is my VS Code, IDE, and you have… this is, like, an entire repo that we work off of. for, like, everything that we’re doing, a bunch of courses, lightning lessons, like, all of that stuff is in our repo. And right now, I’m going to demo you what we’re, like, you know, going to talk about. Okay, so this is my prompt. I’m basically asking it, we just finished analyzing Nova Mart’s 2024 data. So Nova Mart is like a synthetic, database that we created.
you could take it as, like, you know, our, like, we try to recreate Amazon to some extent, like a mini version of it, where it is, like, a fictional company called NovoMerge, and it has all the product-related, you know. issues, metrics, and the funnel, like acquisition to revenue and monetization, all of that part of it, it is… it has 13 tables, 6 million records, and, like, as may… we try to recreate as much real data as pos… real scenarios as possible with this data, right? So, we also use this for all the free workshops to share with you, like, what we could do with this data, with Cloud Code, and we also share this as part of our free repo, that Shane shared.
You could create the synthetic data yourself, and as part of our paid courses as well. We basically run all our Analysis, scales and agents on top of this data. Okay, so, what am I asking it? I’m asking it… this… we have created this analysis, and can you explain me what could be the headline, what are some of the charts that we created, and what is… what are the insights so far? I asked it for a summary, right? And this is what it gave me. So, it gave me… here’s the full readout of NOMART 2024 data. Noomart had a breakout year, revenue grew 5.4x to 3.15 million, but the growth is leaky and getting more expensive to sustain. So, it has some… insight looks like already, right?
So, acquisition scaled fast, and it’s talking about, like, you know, the customers we are paying for con… paying for converting words, and they are not coming back. It’s talking about the metric numbers here. What’s the top line, how many orders, how many customers, what’s AOV and stuff. And it has the top insights. It basically has… this is the revenue, this is the, like, graph, and you know, all of this, right? So, it’s also asking me what to publish. So, it gave me… for a VP one-pager, you could probably publish a… not all of it, but some parts of it, right? It is only picking up few charts. But for slides, it’s also adding a fuller deck.
For Notion, it’s saying, hey, get all the 18 charts that you have that we built. Now, let’s do something. Let me ask it… Let me ask it if it could help me understand, how Claude got here, right? So let me do that. I’m gonna ask for… And… This is how… Can you basically give me a minute… Oops, oops, oops. Okay, can you? shared… with me… I can actually talk to it, probably. an ASCII diagram of all the system skills and agents that Claude Code used to come up with this analysis in the first place. So, what am I asking for?
I’m asking for, basically, what are the bunch of skills and agents that we have in our repo that it used to come up with this analysis, so that you know what is the system behind the system, while this probably Yeah, so this is basically… is sharing with me what exactly was the entire you know, repo structure, the agents, the skills that were used to generate it. It’s still giving me, like, a shorter version of it. I could ask even more detailed version if needed, but if you look at it, there’s an orchestrator, and it’s Cloud Code. It basically plans, delegates, validates, and writes. And it is using all these bunch of things, right? It is having, basically a question framing skill.
It connects the data, and then data profiling, understanding what’s happening, and each of these skills actually invoke a bunch of other skills and agents as well. If we have time, I can share with you by the end. Of the demo. into the details of what type of skills and agents we have here, but I just wanted to share with you, like, the system, the number of skills and agents that we have that helps create an analysis, like this, that gives you, like, you know, the top insights Oops. Like, the top insights and comes up with visualizations and stuff, okay?
Yeah, so it basically does question framing, it connects the data, it does the profiling of the data, and then it looks at… it comes up with an analysis design, which cuts matter, which do not, and then it queries the sub-agents, right? It, we have DuckDB, that’s the database that we have. because it’s local, we try to use that. If you are working with Snowflake or Databricks, you could have MCPs connected to it and have your queries run. on Databricks and Snowflake via MCPs, but for our purposes, we use a DuckDB local database, and we use SQL.
If you see, it used 8 cuts, and it started doing all of that parallelly, and then it uses skills like triangulation, ensuring that there’s some validation and checks, in the type of output and insights we share, because we don’t want to contradict each insight with each other. So these are some checks that it does, and after that. There’s, you know, after the sanity check and validation, and the numbers reconciling with each other, we will look at visualization patterns and create some charts, and then we’ll do the synthesis, and then create multiple documents.
Like, Google Doc Export is a skill like that, we have a bunch of skills for Google Slides and Notion, and finally, we come up with these three styles. Of, you know, like, output. we could do. Awesome. Okay, now, let’s try to get this into a Google Doc. Let’s see. if… it could use this and, you know, get to a Google Doc. So, I’m basically asking it, based on all these insights and charts that you have, could you please create a Google Doc for me? And I’m also talking about the Google Doc being. written for the VP. Like I shared already in my slides, we are trying to create a Google Doc for the VP. And as soon as… We get that… Going, okay.
So, since this is already an analysis that was done, we got this super fast. If this was not an analysis that was already done, we, the cloud code would definitely take longer time. It would definitely take more time to, basically come up with stuff, and, you know, put it together, of, like, the doc, and, you know, put it together… put together some of these, like, charts, and all of that stuff. Okay, let me share… What is the document that it created? Okay, so this is the document that it created. It basically came up with Nomad 2024 product review and Q1 priorities. What’s the recommendation? Noomart grew by so-and-so.
And it talks about mobile is leaking money, which is the number one opportunity. It gives me a deck, a graph out of it, and then we don’t keep the customers within. If you look at it. The title, the graph, all of that, is Has an insight hidden. in it already, like, not hidden, an insight. Basically, the title talks, like, to us with an insight. It’s not generic, like, Q1 customers, over, like, you know, last 20… entire 2024, where… the title is not just a description. The titles of charts, the titles of, Of the, you know, our sections, they are all directly talking to us with the insight. and it basically gave the context enough, and then, it all… it gives, like, a recommendation.
So, because it’s a VP, we are just telling them, hey, this is what we need from you, we want to prioritize this, we want to greenlight this, and we want a decision, right? So. you wouldn’t have something like this for your team, because that’s not what you’d need from your team. What you need from your team is a review of the doc. But for… with a VP, all we want to share is what… where is the insight, and what is the recommendation, and what do we need from them, To ensure that the recommendations that you have get actually done. Okay, so…
Shane Butler: Hey, Sprobi, we had a question around, is the doc already created before verifying the analysis? Or is a doc created after, like, only upon prompting, or is it automated? In the workflow to create the doc.
Sravya Madipalli: Yeah, so… I, today, am sharing with you, once we have the insights, once we have the charts ready, how many styles of analysis documents could we create? But the iteration of the analysis, what do we want to present, the insights, the charts, that’s an entire different set, of, you know, type of work. We’ve shared things like that before in free workshops as well. We call them root cause analysis, where we actually went through The step-by-step process of what are, like, you know, the decisions that you had to make. with Claude Code to ensure that you land on the insight. today, we are taking the inside.
We already landed on the inside, we know what, what charts we want, we, all we need… we’re doing, in today’s is, now that you know what are the places that you could publish this insights, and how do we do that, right? And another very important thing is we’re going to have giveaways for you guys, because you signed up, you came all this while, so, we really appreciate that, and we want to make sure that you have… get some value. And the value that you’re going to get is, I’m gonna share with you the MCP’s, Or, in fact, CLI. I’m going to share with you the command line interface of Google. How do you do this yourself?
I’m going to share with you a PDF on how you connect your Cloud Code with the Google. CLI, so that if you have an insight and analysis, all you need to do is just type the prompts that I’ll share with you, and get the analysis into Google Docs, right? And also Google Slides. Get the deck into the Google Slides. Because, that’s the focus of today’s session, where we are trying to tell you that not only Cloud Code could give you insights within Cloud Code, and you could work with Cloud Code, but you could also take all these insights and publish into a software outside Cloud Code, within MCP, with the CLI. And that’s the intent, yeah. Okay. Awesome.
Attendee, I’m saying, what are the inputs, data files for us to run those prompts? You could get all of those from the free repo, if you’re in… if you could like, you know, try it yourself. We also have a bunch of things. We have a course coming up. This Saturday, Sunday, actually, where we literally walk you through how you get the repo, how do you get it installed, how do you get the data yourself, and, you know, run all these analysis yourself from end to end, but we actually also have all of that incorporated in the free depot, if you could do it yourself, and you know, if you know Claude Code enough, you could definitely go and, you know, try it out yourself as well.
Shane Butler: Yeah, we work with some synthetic data in that, course. We… basically have a local database we build. We also connect to a Snowflake, instance. But you could do anything, right? If you… you can go find a public dataset, you can use your company data set. I’ve been doing a lot of stuff the past week with, Strava, which is, like, a fitness running app. And, they have an MCP to collect their data, so that’s been pretty fun. I’ll share the link again.
Sravya Madipalli: Awesome. And let me type this. I’m going to ask, now, now that we have Google Docs, let’s try Notion, okay? So, the reason why we have all of this is, if you type slash MCP, you’ll get a bunch of, like, MCP servers that are on your cloud code, and you would understand if you have MCPs connected, you would see these MCPs. For example, you see that, I have a bunch of MCPs, and I need to keep validating it every time I try to use them, right? So Notion is something I try to validate, you keep, basically need to go and authenticate, it opens a browser and you authenticate yourself. I did that just before the session, so you have Notion connected here. That’s what it says, right?
I… I don’t have Slack connected, Snowflake connected, yet. for those things, I basically go and, you know, click enter, and then Slack’s-related authentication kicks off, and I need to say yes, and that’s how. you get MCPs connected session to session. First you install MCPs. you go through MCP connections, and later on, in every session, or between sessions, between a couple of sessions, you keep connecting and authenticating with MCPs, right? And today’s session, we are basically trying to show you how can Cloud Code connect with all these different types of software that are existing outside, and share the information that’s sitting in Cloud Code’s brain to others, right?
So, I just wanted to share with you, like, some, like, you know, back, back of the en… like, the structure, or, like, the, frame, how do I say it? The backstage thing that happens. when you have a publishing analysis to Notion, you need to make sure that your MCP with Notion is connected, and mine is connected right now. So let me escape from this, and ask it. Can you help me create a Notion document. I’m currently talking to it. Sorry. Haha. I’m using Whisper Flow to talk to Cloud Code right now, and everything I speak, is going to get into the terminal, and you can see that.
This Notion document is something that I’ll share with my team, and make sure that it contains all the details From executive summary to different parts of the funnel, and also contain, a bunch of… details regarding the different parts of the funnel, and ensure that all 18 charts that you have created for the insights get shared, and make the Notion document as detailed as possible. and have multiple tabs in the Notion document as well, so that we could leverage the multi-page structure of Notion. Okay, so… this is something that I’m asking it to create right now. Let’s see how long this’ll take.
If it’s going to… if this is going to take a really long time, we have around 14, 16 minutes, I want to share with you what it created for me before, right? Because this is something that I ran before, Ein… So, it is continuing it, so in the interest of time, what, if possible, by the end, I can share with you what the, you know, output would be, but let me share with you the Notion doc that it’ll come up with, that I just ran before our session. Ta-dum… Where is it? Okay. Yep. Yeah. Okay, so this is the Notion doc that it got shared, guys. So, if you look at it, I asked for it to give me the full analysis, because this is, like I said, something that we’ll share with our team.
Look at these things that Cloud Code is actually pretty good at leveraging the software’s inbuilt, you know, formatting tweaks, especially with Notion and the New Age document, you know, styles like Notion and Coda. It basically created this little summary blob for us. It has headline metrics, it has story and findings, it really nicely embedded the charts. it has a recommendation, and look at these emojis! I didn’t ask for the emojis, it created the emojis, and, you know, all of that, and… It created all these tabs, and you could click on them to go to these tabs. Look at this, revenue and growth. No March top line in 2024.
What we found, and it gives… it’s giving me the charts related to it. Ideally, I’d like each of these under… the chart under each of these. you know, text, rather than all the text at one go and the charts at one go, right? So this is something that I would work with plot code and ensuring, hey, I like what you have, but can you make sure that the text describing the chart is on top of the chart, but don’t stack the text together and charts together. I would probably give that feedback to Cloud Code if I, you know, was doing this actually real time. And… I go back to here, and then I could look at all other tabs it created for me.
Look at this, Acquisition channel, the so what, and it created the device and the mobile gap. Look at the nice icon it used. And it also has the so what incorporated here. And it talks about the retention, how many… how does it look like for retention? And it also has the categories and margin, what is the so what, now that we have this, and it also has the operations and risk. And the type of, like, you know, graphs it chose to share this with us. It basically talks about what the support load is, the Nomad is being US-centric, and all of that. And the one more cool thing that… oh, wait. That it did, for me, is the appendix. Appendix also has the SQL and methodology.
Ideally, in my regular work setting, we generally give a Databricks notebook link that people can click on to understand, but here, since it ran in the local DuckDB, I made it, you know, share the SQL. With me, that… and it also has methodology, because we want to share this with your team. You want to make sure that you’re querying the right tables, and you’re using the right filters and stuff, right? Awesome. So, that’s where we… so we have 12 minutes. I can literally share with you the Google slide deck as well, and we could go, probably, to answer questions, and I’m sure there are tons of questions that you might have. Ta-da, where am I?
Shane Butler: Yeah, if anyone has questions, feel free to… start dropping them I really like, the Notion one. Also, like, we’ve been talking about, you know, share your analysis everywhere. I actually like it for, Checking it myself, because, like, you know, when you’re working in Cloud Code, like, the output a lot of the time that they’re creating… it’s creating for you is, like, markdown files, which is fine, but it’s like… I don’t know, it’s, like, kind of like… tiring on my eyes to, like, read through those and switch around, versus, like, have a Notion doc where the SQL queries are right under it and everything is right there. And it’s just so fast to be, like.
put this in this other format for me to read. It’s like a… it’s kind of just like a… I don’t know, a UX improvement way.
Sravya Madipalli: Absolutely.
Shane Butler: For reviewing the work myself.
Sravya Madipalli: Yeah. Sometimes clawed code in the terminal is kind of daunting on your eyes, it’s too much information, you get overwhelmed, and I do that a bunch too. Almost anything that I read, I’d like to.
Shane Butler: summer.
Sravya Madipalli: put it in… put it… put the summary in Notion for me. And another thing that I do, like a trick for you, if something that might help you guys, is I comment on my Notion and CodaDocs, so that… and I tell it to answer my comments within those Notion comments as well, so that I don’t have to work with Claude Code and, you know, the details. If it’s a really long conversation, I do it with Cloud Code. If it’s something simple, I try to comment in my docs, and Claude Code reads those comments and gives me, like, an answer to those comments right there. Awesome. So, let me quickly share a similar prompt that I give to it telling that, hey, can you create a Google Slide Deck for me?
This is what it creates for you. It basically has the insight and the graph. Right away, and it picked the most important graphs, and it came up with recommendations. Would this be a Google slide deck I’d share? Probably a slightly different one, maybe. I’ll probably have slightly more text explaining this. What the, you know, what this is about, because you probably need slightly more detail than what it is, maybe a couple of lines underneath. I’ll work with Claude Code to ensure that, hey, the next version of slide deck, could you please add slightly more detail about, these… you know, charts. I didn’t actually iterate much with this, guys. Generally, I try to iterate.
I wanted to show you what the first version of the prompt output would look like, and that’s the reason I have it here. In real world, I probably will iterate and ask the deck to be my perfect version of what I’d like it to be. Okay, cool. Yeah, and wait, let me quickly run through the deck, and you know…
Shane Butler: Can you tell them a little bit about what’s coming up in the next…
Sravya Madipalli: Yes, exactly.
Shane Butler: Jam that, and then also share out the, after that, share out the… Yes. One thing, one question here, does it update the same doc? for iterations, you can do that. You can have it, just depends what you want to do. I… sometimes I have it, like, not update the same doc, because I’m not sure what it’s gonna do, so I have, like. One, like, version… like, two versions of it going at a time. So, like, a backup, basically. Although it’s pretty good if you say, like, undo your… what you just did. But you can give it, you can give it permissions to edit it directly.
Yeah, a lot of the Google stuff is… it’s annoying to get set up, honestly, which is why Savia has created a bunch of instructions around it, but it is nice in terms of, like, I think it’s the most… It’s the one that gives you the most capability, like, hey, you can do this and you can’t do this, like, you can set up, like, hey, you actually can’t edit docs, you can only create copies of them, if you wanted that.
Sravya Madipalli: Awesome. Yeah, I just wanted to quickly run through what we have coming up. We basically have… we completed our first Intro to Workshop session yesterday, we’ll have that in a month from now, I guess. But what’s coming is this, guys. We have a build it, building the system. Remember I shared with you all the skills and agents? We have around 60 skills and, you know, 20 and 30 agents that does this work for you. Get the analysis out the door.
all the iterations of the analysis, how do we work with Cloud Code to get that out, what are the skills and agents that were used to do the analysis, and how do you create those skills and agents yourself, and understand how we build this system, how we build the entire, you know, AI analyst system. We’ll share all of that with you in the bootcamp that’s literally coming this weekend. And we have another version of the bootcamp, which is very advanced version of it, where we basically try to get through multi-agent orchestration. How do you get one agent to verify other agents’ work? And how do we do the auto-research loops?
How do we ensure that, you use open source models, codecs, and the validation piece, and the context piece. How do you get that tighter, so that you basically… not just, you know, help get Claude Code to help you do your work, but actually, like, 10x, 20x, even 30x your work and build systems that helps, your entire team as well, right? So that’s something that we’ll cover in the Advanced Bootcamp. And another parallel track, so the top two ones are majorly about building systems with Cloud Code. But we have another course called AI Analytics for Builders. That is starting next week as well.
You would… you could consider this as a crash course for masters in product data science of Silicon Valley Tech Company, where we literally go through entire framework of what a product data scientist at Meta, at, you know, at Netflix, at all these companies, that we all worked at, and… We share with you the frameworks of how to frame questions, how to define metrics, how do you do deep dives, how do you design experiments, how do you do causal analysis, and how do you drive decisions and actually get impact. All of that would be part of this 5-week course called AI Analytics for Builders. We have some really cool discounts and packages.
We kind of have a bundle where you could, you know, combine basically the, you could combine the bootcamp with the analytics for Builders, and you would get that for $1,800, and we can actually share more with you, but we have coupons, the codes as well, that gets you 20% off of these. Okay, so there are a bunch of giveaways that I’m going to give with you, all. The most important thing I would say is connecting Cloud Code to Google Docs. I’ll be honest, this is part of our paid bootcamp. MCP and CLI connectivity thing, but we decided we’d love to share as many, you know, things that are valuable for you.
So that’s the reason we decided we would share this with you, so that you could start benefiting from connecting with Google Docs today, and getting your analysis shipped there. Awesome. So let’s jump to questions.
Shane Butler: Attendee had a question. Which of the three bootcamps best suits someone that is into analytics engineering and leads a small team? I don’t know what your, opinion is, Trevi. I said, like, it kind of depends on your learning goals. Like, the two bootcamps, the Call Code Dynamics Bootcamp. We need a… we need, I don’t know, a better name for it. And the Advanced AI Analytics Bootcamp. Those are about building… we should just… we should put that in the… In the title, back in the title, we used to have it.
Those are about building, and then Agent Tech Analytic Systems, and then the five-week thing is about doing analysis, so it’s like… using agentic systems to do the analysis, so it kind of depends, like. For your team, like, oh, are you focused on, building… a Genetic Analytics Engineering team, a system, then maybe that’s more like the bootcamps. But if your team’s also going to be, like, doing end-to-end Analysis of the data, then that’s the 5-week.
Sravya Madipalli: So, one thing I’d like to share to what Shane added was… We have designers, we have, like, salespeople, we have product managers. Like, all multiple different styles of roles, taking all of our courses. They… in fact, people who are not in data, or people who are adjacent to data science would benefit a lot from AI Analytics for Builders, because they literally understand the entire frameworks of how product data science works, right? And we also have a bunch of product data scientists who took the course themselves, because that’ll basically help them understand their… the frameworks that they’re used to, all their carrier, but how do you use those frameworks with AI?
Because we teach not just the frameworks, we teach how do you use those frameworks with AI. So that’s what the Builder’s Course is about, and it’s a great refresher, and a great, AI companion for, like, how do you use frameworks with AI companionship. for data folk, for non-data folks, it just gets you, opens the door for, like, doing anything with data. So that’s the reason we… it’s been popular in, like, non-data people as well. And coming to the agentic… building agentic systems, the two bootcamps, those are for anyone who would like to basically build and work with agentic systems and 10x your productivity thing.
And even more beneficial if you’re actually a data person and you’d like to use the repo that we built, the AI Analyst Plus repo that we share as part of the bootcamp, where it contains, like, multiple myriads of, like, skills and agents that goes from, how do you frame a question better, what type of visualization do you want, how do you like the colors to be, the themes to be, what should be the title, like, you know? Two, actually, things like, how do you design a metric? How do you validate metrics, and stuff like that, yeah.
Shane Butler: Yeah, and actually, maybe to even speak a little more, like, kind of, like. philosophically around, like, why we created these courses. Well, particularly, actually, just the AI In Lunch for Builders, the 5-week course, is, like, we are really trying to prepare people for where we believe, like, the puck is heading. And so, what this came out of is we have this podcast, and we interviewed about 10 CEOs, founders, VPs of a GenTech Analytics Company, like, you know, like, the CEO of Hex was on our podcast, the VP of Gen AI for analytics of Tableau was on our podcast, and this was back in Q3 of last year, and it came very apparent to us that, like, hey.
Agentic Analytics is, like, this is coming, for sure. There’s so many people in this market who are creating this, and this is before, like, Opus 4.6 and stuff, it’s already pretty good. And so… We saw this… we’re envisioning that, like, you know, more people data scientists, data analysts, they’re going to be able to do faster, more robust, more analysis on their own. And then other people who are kind of, like, tangential, maybe they’re upstream in the workflow or downstream, they are going to be able to go more end-to-end and leverage data in their own workflows. So, like Savia said, designers and PMs, but also analytics engineers.
So, like, if I think about it in terms of, like, what I’ve seen with, like, my team that’s worked on it is that Okay, data scientists, data analysts, they’re able to get a bunch more analysis done. Now they have free time. what are they gonna do with that free time? Well, they can do a couple things. They can go more end-to-end, but there’s two sides to that. They can either now act on the insights and start doing more, kind of like, product development work that the decisions from their analysis are making.
Or, they can start doing analysis that they couldn’t do before, because the data wasn’t in a way… in a format or available, the data foundations weren’t built, so they can start doing some analytics engineering work. Not great analytics engineering work, but they can stretch to that, because it is just upstream of them. Same thing with the AEs on my team. Is… now, they are, like. making our data foundations better than ever, because they’re building ejectic systems to, you know, build their dbt models, and so they’re… they’re already almost doing the analysis as well, and some analytics engineers, like, already do a lot of analysis.
The ones we… I work with, they’re not quite doing it, but they’re working with the stakeholders, they’re building the metric definitions. And so now they’re able to, like, take it that one step further into, like, actually doing, like, data science and analysis. those are… so I do think that 5-week course, if your goal is, like, to stretch teams into, like, different types of domains of work. That is an opportunity there, for sure. That’s why we created it, at least.
Sravya Madipalli: Yeah. I know we’re out of time, but we could take a couple questions, or something. If you have anything you could ask live.
Shane Butler: And I think we have some stuff we’re gonna share out. Are you gonna share that out, in an email later?
Sravya Madipalli: Yes, I would share the email. You’ll also receive the recording from Maven as well, so for, like, you know, later on, if you want to, like. follow up, I totally suggest you to, like, try the Google CLI connector. You would be, if you don’t already have done that with, like, you know, working with MCPs or connectors. The first few times is definitely magical. I remember it from 6 months ago, or… I don’t remember now how long ago, but it was… it definitely changes how you work.
Shane Butler: So I dropped the builder’s course in the chat there. Oh, do you have a promo code in your slides for the… Bootcamp for this?
Sravya Madipalli: Sorry? Yeah, I… the promo code? Oh, I can give the promo code, yes, yes. I’ll share it in email as well.
Shane Butler: Okay, cool.
Sravya Madipalli: Attendee has a question, do you want to take that, Shane, while I get the promo codes?
Shane Butler: Yeah, so there’s, okay, so we run the boot camp. We’re gonna try… we’re trying something new, actually, this time. So, typically, we run at 7am to 11 a.m. Pacific time, Saturday and Sunday. So, 4 hours… in the morning, it’s all live in Zoom. The reason the boot camp? is live, is because the… And the space is changing so fast. I don’t really want to record videos about how to do something in Cloud Code, and then it changes two months later, so we’re constantly updating the bootcamp. Whereas the 5-week course, it’s part async, part live, there’s a lot of evergreen information, or like, hey, how do you do root cause analysis? That’s not gonna change. But for the boot camp, yeah, 7AM to… 11 a.m.
Pacific time, but then we are gonna have some… We’re gonna try something new this time, we’re gonna have a bunch of bonus async material, around validation, more data warehouse connections around context management. That will basically have Monday, Tuesday, Wednesday. Thursday, the following week, that you can do on your own time. It’ll be like a 20-minute video a day, some exercises, and then you can kind of let it sink in through the week, and then Friday. I don’t know what time, probably the morning again? Pacific time, we’ll do a office hours that’s optional, and kind of sync with everyone and see how the week went after doing the bootcamp.
And then the other… so, we’re experimenting with that, and then the other thing we’re going to experiment with in July, if you don’t want to do a weekend, or if this weekend’s too soon, is we’re going to do a weekday version of it, where we do Monday to Friday, 2 hours, 7 a.m. to 9am Pacific time. Each day, that’ll be recorded. The recordings will go out. I’ll add an hour later, so around 10 a.m. Pacific, if you want to do it async. So, just to add some flexibility. So, I’m not sure exactly what that time is in Europe, 7 AM, so, like. What is it, like, an 8-hour difference, or 11-hour difference, or something?
Sravya Madipalli: I think it is, 7AM is around 3.30 PM?
Shane Butler: Something like that. Yeah, so it’d be, like, your afternoons.
Sravya Madipalli: We’ll take a beam.
Shane Butler: Yeah, and then I think the weekday version will be interesting, too, because then, you know, you can, like, take 2 hours in the morning. Go try stuff out the rest of the day on your own. Do some exercises, do it at work. And then you can come back with questions the next day, doing their 2 hours, and it’s not like… As intense session for 4 hours, but some people like more of intense, so we’ll try them both out. We are definitely up for trying other stuff out in the future, too. I know we don’t… we don’t… we’re not as Aussie-friendly as we should be. We need to do some evening courses, Pacific time, so our friends in Australia and New Zealand have a better time. Cool.
Well, yeah, two for one… With the 5-weekend, the boot camp coming up, if you’re interested in that, basically give us a DM on… on LinkedIn, or… or in the Slack channel, or we’ll send out an email, and you can reply there. You can… Email me at… Shane at Aynoslab.ai, anytime, if you wanna… get on a call for 15 minutes and chat about what’s best for you, just let me know. I’m pretty free.
Sravya Madipalli: And I just shared the links with promo codes as well. So that you all are, can, like, you know, access them, yeah.
Shane Butler: I don’t know, anyone not in the Slack? I feel like most people are probably in the Slack, because I recognize a lot of the names here. But, we do most updates there, so you’re not in Slack. There’s the link there?
Sravya Madipalli: Awesome.
Shane Butler: Cool. Any other questions? Nice.
Sravya Madipalli: Okay, this was great, a nice cozy session. Thank you so much, everyone. I’ll share with you all the giveaways soon, yeah. See ya.
Shane Butler: See ya.