AI Rebels

Plotly Thickens — How Open Source Is Powering the Next AI Wave ft. Domenic Ravita

• Jacob and Spencer • Season 4 • Episode 4

Share episode

What if rebellion looked like collaboration instead of chaos? Plotly’s VP of Marketing Dominic Revita traces his journey from coding the first internet-only bank in the 90s to shaping the next generation of AI-native tools. He reveals how open source innovation and AI-driven creativity are merging to make data storytelling more human—and more powerful. The conversation dives into the philosophy behind building transparent tech that empowers rather than replaces. It’s a visionary look at the future of AI as a partner in progress, not a threat to it.
https://plotly.com/

00:00 – Welcome to AI Rebels
Setting the tone: why innovation and rebellion matter in AI.

00:45 – Meet Domenic Revita (Plotly VP of Marketing)
From engineer to marketer: bridging tech, data, and storytelling.

03:00 – From the Dot-Com Era to the AI Age
How Domenic went from coding the first internet-only bank to building modern data platforms.

08:30 – Lessons from TIBCO and SingleStore
Real-time analytics, distributed systems, and the rise of product-led growth.

13:45 – Inside Plotly’s Vision
How open source and interactivity changed the way we visualize data.

20:00 – The AI Shift: From Tools to Teammates
Why the future of data analysis is agentic, automated, and deeply human-centered.

27:30 – Behind Plotly Studio
Exploring their AI-native app that turns raw data into insights in minutes.

35:00 – Serving the Next Billion Knowledge Workers
Empowering everyday users, not just data scientists.

40:00 – The Power (and Future) of Open Source
How open ecosystems are outpacing closed models like “closed AI.”

46:00 – The Future of AI Collaboration
Where human creativity meets intelligent systems — and what’s next.

49:00 – Final Reflections
Domenic’s advice for builders in the age of AI innovation.

hello everyone and welcome back to another episode of the AI Rebels podcast as always I am your co host Spencer and I'm your co host Jacob and we're here with Dominic Revita who we are very excited to have on the show from plotly and many other adventures that we'll maybe touch on thanks for coming on the show Dominic hey happy to be here guys now Dominic you have like I mentioned a varied background you're currently the VP of marketing at plotly you're kind of this unique blend of engineer CTO marketing you've been involved in multiple very successful companies now you're involved with AI directly I'm curious can you tell me a little bit about your journey how did you get here to plotly sure yeah I um I started as being a developer back in the 90s so it's been a long path since then it was a huge wave the dot com wave at the time I was writing C++ code for the US's first internet only bank we were called Security First Network Bank and probably one of the first five in the world but the first in the US in the in the mid late 90s and it was amazing working in that environment every day was an exciting kind of new day and I had come from you know developing C programs against Sybase databases and things like that so I was starting to do the new thing which was object oriented programming in C++ and it was just fortunate to be able to work at a place like that one thing LED to another I continued my developer career moved on to Java then I joined a company that I stayed with for 16 years Tipco Software Tipco is an acronym that stood for the Information Bus Company so they were real time data distribution and they started on Wall Street to do real time speeds and feeds for equity trading and exchanges uh the company expanded into the full software stack business process data integration complex event processing visual analytics and data science over that time so I was um a solution engineer in the field so I'm hands on with with our products and systems with our customers hmm designing systems really solution architectures and I got exposed to a lot of different ways of technologies and b to B data infrastructure and analytics over those years then I went back to I I did a startup earlier in in travel technology just before I joined Tipco and you know a few years back I decided I'd go back to the startup world um and I joined brings you back haha yeah it brings brought me back and I became a field CEO to a company called a company called MIM Sequel now called Single Store a distributed analytics company and that is sort of the step like towards marketing in that field CTO role hmm and I start I took on product marketing and that one thing LED to another and it was adding more you know responsibilities in marketing and that eventually LED me to the head of marketing at at plotly interesting tell me a little bit about like what did you learn at Single Store that you're now applying it plotly is there any big lessons there yeah that was the big lesson in growth and product LED growth in particular all the mechanics of how do you build this how do you you know what does the product have to have what's needed from marketing it's a it's a very cross functional collaboration between product engineering user experience design mm hmm um and marketing right because in traditional sales and growth you know you have the website and you have content and all these other assets that are the ways your content sells for you in a way and with PLG it's the product doing the selling for you so all of those similar ideas you would have about engagement and how do people learn and educate how they know what this thing can do you translate those many of them not all of them but many of them right into the product functionality so those things become like first class citizens inside the product you wanna keep people in the product as much as possible and a single store we were doing this with you know basically their their product is a real time analytics database yes so how do you do that like well you need people to build apps on top of that and but they need to understand what's different about their app yeah I could just use Postgress why would I use something like this so there's a it's um there's a lot of fast iterations and I would say there are a lot of lessons I would take away from single store but there's there's just that product fast integration is one interesting secondly is the the economics of it right because you have to think about what's the cost of the usage because it's it's a SAS product so you're paying for the usage in the cloud on AWS do you have the right margin yeah so it does bring the tactics closer a lot closer to the operations like cloud operations and engineering interesting um so with the the real time data analytics that seems like a um seems like you guys would have been pretty early in that space so I'd be curious to hear if if that's kind of a if you bring a similar mindset to you know this this AI space now that you know it's starting to take off more yeah I well gosh AI I've been working with various kinds of AI for 15 years and as Tipco moved into the analytic space we also moved into you know data science and and AI enabled capabilities in that in that time so my view of AI is sort of that some of these things are the same things that we've been doing LLM of course are totally new in the last few years the generative AI part of this but other sub areas of AI you know we've we've seen and we've been doing and it's always about do you have the right data first of all to do classical machine learning in fact right clean data you need clean data you can do the feature engineering etcetera etcetera there's still a huge amount of mileage out of those other AI techniques not just generative AI so when I think about real time analytics and what we were doing back at single store to me that was also an evolutionary step yeah because there have been ways to to do when we think about real time analytics it's like how can I process X amount of data to make X decisions in X time frame right and yeah serve serve those responses and like what is the time frame for each of those things and there have been ways to do it in earlier years like back in 2005 and 6 we were doing complex event processing and stream processing so today those technologies have emerged or evolved into things like Apache Flink you know yeah uh but the way we looked at it at single stores like well all of those tools were sort of very limited could only answer a few types of questions do simple aggregates sum medians averages if I want to do a more complex calculation on terabyte pipes of data continuously or many calls and have lots of people access it really I need a full fledged database with all of its features and all the expressive capabilities of SQL and uh so I would say in that in their case that single store is a simpler way to do streaming analytics than a lot of these niche things like flink or materialize or things like that and and to me that's what's changed in the database space is distributed databases you have specialized types for each of these different types of workloads yeah and so there's a policy for AI there too yeah definitely wow okay so Tipco single store so what was it that then pulled you to plotly what did you see what did you identify that you were like okay this it's worth the a job change is big I just went through one four months ago and it's big it's a big decision so I'm curious what what LED to that yeah I well first of all plotly is open source and yeah I think the business model of today uh the most successful ones in technology have an open source element as part of the business model and the reason that's important is because you can collaborate and build an audience with open source interesting and you can get a ton of feedback earlier about what is working and it's hard to really get that reach any other way in especially in b to B infrastructure when you're selling to developers data scientists analysts there's just so much free open source it's also a proving ground like is your project worthy enough is it valuable enough for people to use when everything they could choose five other free open source things why would they choose your thing so that's number one is that I was looking for a company that had open source as a key component of their business model secondly I was looking for a company that was in the right stage for what I wanted to do next which was basically repeat the early product market fit to yeah scale at you know tens to hundreds of millions in revenue having done that at single store through that phase I thought okay let's let's do this again in another market I'm familiar with um in this case an adjacent market with data analytics at Plotly um and it will be similar but different different set of constraints different but the the growth path and the same obstacles right you know you need to get past so it was a it was a good fit for that the third was what's a company that want that wants a marketing leader with an engineering background and that domain expertise not all companies really want that so I was looking for something really specific where a company that wanted that combined background of hands on with this kind of product the sales experience through solutions engineering for all that time at Tipco and then the go to market experience of how to take that and put it together for their audience there was a fit on that and also forth the people are great the culture is amazing we're a Canadian company based in Montreal and we're fully remote but we have offices in Vancouver British Columbia and Montreal and yeah just uh having worked for Silicon Valley companies and New York based companies it's a totally different vibe really in what way I I just find that people are really friendly haha yeah you know you can say that about Canadian companies but it's I think it's it's it's true there's a lot of flexibility by being remote uh you know there's you have to like communication you have to do some extra things with remote companies like that to keep people in touch but it really works in terms of hiring talent because we can get people you know yeah pretty basically for sure time zone you want certain time zones but uh it gives us a lot of flexibility on hiring yeah OK um no that makes sense so can we dig into plotly what's yeah what's what's the big idea what's like the 100,000 foot view what's the mission what's the the goal yeah so our co founder Chris Parmer when he and his co founders really created the company they were they wanted to solve this problem they were you know um scientific researchers data scientists and what they were working in different fields like like biology and um they wanted an easier more flexible way to do interactive data visualizations and they just felt that like what was available just didn't solve the problem there were all kinds of restrictions there were proprietary tools that sort of LED you sort of into a black box and so that became the beginning of plotly data visualization and it's interactive so if you're a Python user you can just pip install plotly and that gets you the interactive data visualizations that was the beginning that was the first product we now have over 1.3 billion downloads of that oh my god it's growing by millions per day um the second innovation was you know I can build applications that embed these interactive data visualizations increasingly I mean serving data scientists primarily these data and now AI teams were looking for ways to build interactive analytic applications because they were finding that like traditional business intelligence tools were limited in this way so yeah what is an analytic application how is it different than just a dashboard you know dashboards that you get from Power BI are very good at sort of descriptive analytics and it's typically read only I know the questions here's what it is I'll give it to my users when there's a need for a custom analytic app or we call it a data app for short I'm looking as a user or a data team something that's more read right I'm reading data but I want to also input data I want to interact with a machine learning model I want these days I might have a generative AI interface as well and I'm I'm trying to make this a decision and act on it in the moment so that is sort of part of what the need for a custom data app but also we're finding increasingly the teams who recognize the business opportunities to create these are the data science teams or the data in act they so they're acting it like as product managers so we recognized the need and this is going back a few years now like well how do you enable a data scientist to create an interactive web application with their data without having to learn to be a front end developer and learn all this Javascript and stuff or become a back end developer and learn all of that but they can as you know lower the barrier to creating this and so that insight is what's really motivated Chris to create plotly dash and plotly dash is the company's really next big open source innovation which is a Python based interactive data application framework so Pip install dash and now you can start creating an interactive web app or data app as we call it so an example would be like so UK Power Networks is like a an electrical grid uh operator and energy provider in the southwest of England and they have network technicians field technicians in the field and they're looking to basically optimize customer experience for predictive maintenance but they're also like wanting to minimize their cost of these things so the data science team is the one who has created the models and has the idea for like if we give this app to our technicians they'll have the right information at the right time in the field they can in they can give us more data and we can have this virtuous cycle of improvement with every interaction so they have a series of of plotly dash applications for internal operations but also for customer facing operations that are on the internet for their users to have access to you know understand when they can contribute to the grid with their own solar panels or when they want to cartel that OK and what role would the I know you mentioned the read versus read write what kind of an impact does that have like in this example that you're giving why is that significant here yeah that's significant because you you want to in their case they want these field technicians to be able to to enter data and do what if scenario planning and with the live the live data with the live data and using the models the machine learning models that the data science team has created so that's not something they could create in a traditional BI tool in like Power BI or Tableau now if you did a software project or application from scratch you could do that but it's more effort right yeah so plotly dash serves sort of that middle ground between the two that enables a data scientist to create this kind of this data app in pure Python and easily get the user interaction they need get the collaboration features they need get the right kind of user controls these interfaces tend to be very custom to the use case and so that's very important as well we have customers who are AI GPU creators I'll say it like that there's a few of them and they're very particular about what their their branding is and so one of these customers has created basically a a partner supply chain application and so they use it as well as their partners across the supply chain to monitor what's happening and they want that exact branding and just the user controls and just the data that they want to expose to each of those users so it is a custom data app in that way and again in their case the data science team creating this kind of data product so plotly dash and our commercial product Dash enterprise is a really good fit for that scenario interesting so what how does it compare to like a a platform like Mixpanel targeting similar markets or is it slightly different in scope that's a good question so Mixpanel is an what I would call an in product analytics tool like amplitude or posthog or something like that right so right that's that's a specific use case typically product managers are using this tool and product operations right to monitor whatever it's my users use what across my SAS product you could absolutely create an in a product in an analytics tool like that out of plotly dash and so that's the difference is that Plotly Dash allows you to create any kind of analytical app like that or data app I see okay rather than just an out of the box one specific to a particular use case yeah and that's always sort of like the buy versus build decisions for right yeah any of any application enterprise if there's something off the shelf that does exactly what I need and I just plug it up boom I'm gonna use it as long as it's affordable for my usage but a lot of these custom data apps that our customers are implementing there is no app for the thing because their operational scenario is too specific and or they have a lot more customization to do they're sort of owning their own AI which is a growing trend and that they don't want the AI externalized they own their data and they want to own the AI operations on that data so they're gonna have to build a custom data app and house for that kind of thing interesting interesting so you've kind of hinted at it but what role where does AI play a role in all of this that plotly's trying to accomplish where does are you training your own models is that what you're enabling customers to do is train models on their data I'm curious where yeah so so our like like I mentioned Chris's original vision for Plotly Dash was to make it easier for data scientists to create these apps that would solve the read write problems or to solve this last mile gap without them having to learn all this extra technology well AI like it's doing to a lot of different areas is just completely changed what's possible yeah right yeah and so what is it what is it what does the world look like when AI can start to analyze data for you even in a basic way and if if you tried it with Chat GPT you can sort of see like oh plot this is a bar chart it'll create it's a if it's a small simple set data set it may do it correctly with larger data it might be off but you're like oh you can see where this is going also AI is great at generating code if you give it the right context right and so the main way we're using AI today is is exactly that leveraging the power of AI code generation along with it's rudimentary ability to do some data analysis marrying those two things with 10 years of plotly experience in creating these custom data apps for for hundreds of customers and how visualizations are used and basically taking all that putting it together in an AI native tool we call Plotly Studio which allows you to basically vibe analyze your data I like it I like it yeah you just in you just ingest the data set and it presents and asks questions along with with answering them with visualizations for things that you probably haven't thought of about your dataset interesting so that's how we are using AI to basically up level what's possible for our same audience data scientists but now with this it's so easy to use because literally it's a dataset to an insight in a in about two minutes that it sort of changes the nature of of the engagement with data yeah that's incredible that's big yeah so just plain English prompts well you don't even need a prompt even you can literally just give a CSV or a parquet file and it'll just analyze it for you and provide it it'll analyze it and visualize it and give you the narrative and it goes further than that it generates natural language specifications which now you can modify to remix or regenerate on a component by component basis of the app is there anything you've had to do to kind of engineer the AI to make sure that it's always providing you know salient what's the word salient takeaways from the data or is it you've just found that you know the AI is good enough now that it needs very little guidance to presenting salient points of data yeah it's a good question there are lots of things that we've had to do to to guide the AI and to provide guardrails and context so there's a ton of context engineering that has gone into this um but the thing that it adds uniquely is this world knowledge in the LLM so when we give it a dataset even without a prompt and it's a let's just say it's a series of you know floating point numbers we and we've done we test this daily with hundreds of data sets of various shapes sizes and we're we've we sort of understand like we see where the boundary or the frontier is where it can recognize the data even with no labels and it comes pretty because it's recognizing oh this series of floating points this looks like you know this is mass body index readings it can identify that these look like precipitation measurements and it will give us the words to do that so that's sort of was a really surprising thing in the early development that our engineers sort of came to see that like wow this is a great starting point for how we basically explore data it's a new way to do exploratory data analysis and what we're providing in that first two minutes in that first generation is a completely functional interactive data app and I can put if I like what I on the first shot I can hit one button and deploy it to the web to platy cloud but often you want to tweak and modify it but now you're starting with a much higher sort of turning point ask me questions yeah probably hadn't thought of before right no I love that cause I and I I've talked about this a lot recently on the podcast for whatever reason um but I'm a I'm a web developer by training so something that I have been very interested in from the very beginning is the capabilities of language models to generate user interfaces on the fly and I always thought that that app statistical applications like yours would be the first to to realize this dream of mine so it's really exciting to see it happen it just makes a lot of sense you know you got a lot of unstructured data and LMS are really good at unstructured data that's really fascinating that it can find those patterns even in unlabeled data which like makes yeah sense like watch it an LLM right it's built we're seeing this people I think realize the capabilities of not just AI in general that like pattern recognition we've had several guests on from the medical field where they're realizing oh wow it can go view 100 million photos of various types of cancer which make it incredibly good at recognizing what it looks like when we give it new scans it's just it's the same thing here with data analysis like it this is really what it's built for this is it's bread and butter yeah I it it does help that we have 10 years of plotly open source code out there yeah that's huge it's been indexed and trained right and you know we're getting 5 million or more downloads every day so we're wow we're seeing like the our plotly open source code is being used everywhere you see it embedded everywhere it's embedded in Julius it's embedded in retool it's embedded in data bricks wow plotly might might be like the world's most popular interactive data visualization library out there and so that's sort of been a fortunate you know situation when LMS came on the scene we started recognizing actually it can produce some plotly code that's not so bad but it but it won't only do the basic stuff and that's where we saw the opportunity to use the LLM as a building block and yeah build an what we are calling an AI native product on top of that yes so you you partially answered my question already I think but when you guys are generating these apps for people does the LLM write most of the code within the plotly uh package is it all plotly package code or is it you know supplemented with with uh some extra Python that it knows it needs to add yeah good question so it it generates all the components and the components are visualizations with their filters and interactive controls and it also generates these components which call specifications which is a natural language and it's really just plain English like okay this is a range slider that does this and has this range so if you want to change it go yeah that's great but I want to modify it you just copy and paste that text and regenerate just that component or just that filter and it also provides the code in yeah in another tab that code includes you know plotly data visualization includes plotly dash code but it does include like boilerplate other libraries like pandas for instance yeah and numpai and yeah you know things that you would typically use so and then you know we do also have some libraries around layout and all sorts of other things that help to give a good first experience in that first generation of the app yeah huh another thing I think it's like maybe important for people to understand is like that why wouldn't you do this with just chat GPT or something like lovable that's good yeah yeah like like I and I think what you're starting to see is like or cursor even right mm hmm you're starting to see that the general purpose tools that are meant to be everything to everyone aren't doing anything specific really great like even lovable is like like cursors like I'm gonna help software engineers build any kind of project but that's for software engineers if I'm a front end designer maybe cursor is not that helpful but lovable is because it's more more oriented to designers and front end developers and the things that they're doing more often are like things like create a website and lovable seems to be zeroing in on OK let me create help create a website right right and oh now I'm gonna help you generate the SEO I'm like OK so they are going down the specialization route in our own experience I mean of course we use some of these tools in house is that well where's the tool for AI power data analysis that really you know 10 x is my productivity helps me get to insights faster helps me ask questions better of my audience helps me understand the data better and we didn't really see any good examples of that and so our our view of this next era of the AI let's call it the AI powered marketplace mm hmm is that you're gonna see more specialized tools based on the job to be done and and the user and that's where we where Plotly Studio fits is Agenic Analytics or some people are calling this you know vibe analysis like help me remove all the friction so I can just I'm not wrench turning as as one of our uh community users called it I'm just thinking about what's the question what is this data trying to tell me interesting yeah so when you say agentic analysis by agenic do you mean it's agenic in the sense that you upload the dataset and then this AI agent will perform all the analysis for you or is there more to it than that um that's the gist of it so what happens under the covers is well you upload a dataset so you're presented just with a screen with a button upload your data and you can give it a prompt if you want optionally if you don't you just the next step is you push a button that generates a plan which is basically an analysis plan we call it the outline and it and that's really now where a multi under the covers a multi agent system is invoked now the user doesn't see any of this and this is what makes it so easy but what's happening is in parallel a series of specialized agents are being fired off for a data prep agent and profiling let me profile the data how sparse is this data should are these numbers should they be floating point numbers um are they strings etcetera then there's the the prep do I need to shape this data to help with aggregations it's this time series data the next is a data visualization agent and then an application component agent so all of these agents start to work wow to make decisions autonomously based on what it's seeing in the data with our own context engineering we've provided also the world knowledge of the LLM and that's how it comes to this first you know 2 minute version of your app with everything accessible and modularized and componentized with the spec for each component and the and the code so that's why you know we don't really market it that way though because we're not the problem we're we're not trying to make another software engineering project for our users right there are lots of good ways to go do that with Google's AI Agent SDK or Microsoft everybody there's lots of AI agent SDKs we're trying to keep it super simple it's for the people basically the 1 billion knowledge workers who need to work with data in some way and need instant insight this is who we're serving mm hmm it's powerful I can't believe you said over a billion users billion downloads well we've had 1.3 billion downloads of Plotly Code but the audience that we're designing for now with Plotly Studio is the 1 billion knowledge workers with several different estimates is sort of what counts you know as knowledge works if you work at a computer you know if for any job at some point you're dealing with some sort of data that you need to plan against or analyze even if some I mean most most of those people today use some sort of spreadsheet to do that at the most basic level right they're using I mean I was gonna excel yeah my my background's in accounting and finance and I was in I was a public accountant doing auditing not too long ago four months ago like I said job change but the thought of this just with my background I I'm envisioning being able to upload Excel based financial statements for a company or other data like that for auditing purposes and it just will run through say this is this is not consistent make sure to look at this this is risky the riskier area because X y Z I mean this has so many use cases yeah and it's it's it's really phenomenal we're getting more and more use cases uh and more and more interesting datasets out there it's it can spot trends quicker than you can sort of really explore it in a gooey interface so we're having power BI experts in in that community come to us and like say first of all Plotly's visuals are amazing it would have taken me x many hours to do this parallel line coordinate plot or you know this trend analysis and to to customize it with a thicker line for this trend and color and etcetera and you just did it for me in 2 minutes flat just with wild that kind of customization you do need to find the words in your prompt to do it but if you know how to describe it in your existing tool it most of the time works in Plotly Studio it's I haven't seen many accounting data sets I would love to see you upload some accounting be interesting see what it identifies yeah yeah yeah so I was that that kind of leads to a question that I had which is so you've talked a lot about serving knowledge workers I'd be curious to hear do you have any notable and and from what you've described it sounds like mostly the the people who are you know hyper engaged in your community and in the open source project itself are all hardcore data scientists guys I'd I'd be curious to hear have you seen any use cases that um users or companies that deploy your product that that target a slightly more lay user you mentioned that that Julius uses plotly in in their visualizations I'm just curious to hear if there's any other similar use cases yeah traditionally we haven't targeted sort of delay user yeah we were targeting before Plotly Studio our audience was focused on data scientists data and AI teams building these custom data apps um and so our customers you can see from the website companies like SMP Global yeah I've mentioned UK Power Networks Intuit Uniper etcetera creating these apps in all these different operational contexts to to basically take data science out of the lab and put it into action for the business through custom data out so we're still doing that um and it's a large part of our business but with the introduction of Plotly Studio which is it's in early access right now we see an expansion to add data analysts the just yeah the ordinary data analyst and also the lay person who doesn't maybe think of themselves as a data as a data analyst but from time to time they wanna explore and analyze a dataset yeah right yeah and so that's that's a new audience for us with the introduction of Plotly Studio so I don't all I don't have existing customers cause we haven't launched this as generally available yet but we do have we have a lot of a growing set of people in the early access program who are sharing their results with it so I'd point you to people like um Brian Julius and Art Tinic who are publishing sort of and you can find on LinkedIn under the hashtag Platy Studio a lot of different variations of their analysis with with Polly and they're coming from um you know some of these examples are recreating things they did in Power BI and comparing the differences others is like like net new kinds of work flows like Brian Julius is enabling you know starting out in like R code as the AI coding IDE generating a set of data or a set of code using that code as a input prompt to Plotly Studio to create yet some other like second level derivative that we as of the product creators hadn't even envisioned which is amazing that's why you build in the open and why you do open source like this because it's just it accelerates the innovation really yeah yeah that's that's why I was asked wanted to ask about you know non technical users as well cause I think that it's a pattern that we're seeing across a lot of companies is that they're they're they're starting to find that the non technical users engage with their tools slightly differently um just really interested yeah I'll see where it all goes that we came across because of this new type of user in our early access program is um a standard about that talks about constraining visual communication language yeah so it's basically um the the the founder Jorgen likes to talk about the analogy of music notation to uh visualizations data visualizations right like hey we've standardized around how to notate music for every different instrument you know hundreds of years ago in western music anyway right why can't we do that with analytics in business or data visualization in business it makes a lot of sense shouldn't there be a visual vocabulary that has some consistency totally so that you can measure and compare apples to apples effectively no matter what technology you're using so that that group is the International Business Communication Standards Group IBCS and so we weren't really that sort of thing wasn't on our radar but through the early access usage people started saying hey this I can do IBCS compliance visualizations with Plotly studio hmm and you can do that today and like artinick is one of the the EA early users that you'll find his stuff out there where he's created a series of IBCS compliant visuals and we've put almost no effort into it so far it's just that plotly visualizations are that flexible to begin with yeah and the information about what the standard requires is also public and it's been indexed by the LM that said we are now we're in touch with them and we are gonna be collaborating with them to go the extra mile to to make this to make this work in a consistent manner according to the standard so we're really excited about it but it's that was like a totally net new uh area that that's interesting came to us through this yeah which there's this thread throughout I feel like it's been brought up multiple times is the importance of open source in this innovative path for plotly I I'm curious one I'm a big proponent of open source as well but I just have so many questions because I don't there are not many people who think an open source company can be successful they they worry that if it's just open source anyone can take it anyone can use it how is plotly become so successful how are you generating generating money how how are you maneuvering this really funny world of open source yeah it's I think the answer to that for companies has changed over time right there was not really any successful open source company until Red Hat to me that was the first successful open source company and for years we were asking ourselves as an industry I'd say in the b to B software space anyway who's the next Red Hat who could do that uh could you have an open source database could you have and this is like 20 years ago now right could you have an open source CRM could you have open source this and that and a lot of people at those times are like oh no it's there's no way databases are too complex and difficult and require too much effort and you won't get enough contribution fast forward now we see like yeah databases Cockroach Labs you know tidy B um postgress has been open source since its invention by Stonebreaker decades and decades ago and now it's getting now it's the world's most popular database um and there are all sorts of forks and variants and branches of it so it's but open source has changed and it means you know I think there's different aspects to think about but if you think about your business model which I think gets to your question like how do you make money if it's open source mm hmm and generally I mean our business model is an open core model so what that means is that we provide a a core product plotly dash open source and plotly data visualization for free use it's MIT licensed meaning that it's the most permissible and you can do whatever you like with that and then the commercial product builds and extends it and generally provides the functionality you need in the enterprise context so for us that means how do I provide security for this Python interactive data app right how do I integrate with CICD and Devops how do I improve the performance how can I create pixel perfect pdfs all these things eventually you could build yourself given enough time and motivation but if you want to have you know a set of interactive data apps like like UK Power Networks has I mean there's several internal several external you also want a central place to host them like an internal portal you want a gallery so that other teams don't recreate what you've already done like hey just build another data app and host it here in the app gallery so there's a lot of these sorts of things that that we're doing for our customers but generally it follows this this open core business model that makes sense it just I think that's the path forward that's how innovation will happen that's the true innovation I just don't think a closed model can pivot nearly as fast and ironically open AI despite its name well no you know it's not very open they just released this model and it's debatable what open means in that context yeah would you say that's open if you look at the pure open source definition no it's not like I don't have everything to to create this that's not to say it's not useful and it it does help me now to run you know a model like that you know locally with Olama or vllm or something like that so it's it's useful it's just it's just let's not say it's really open source in the purest sense I'm not sure how much that that matters though depending on what your usage is if you're creating a new model you need to think of like what's the training set what are the weights gonna be all of that so obviously they're protecting sort of their intellectual property there by not exposing all of that but most but you know it's the real thing they're doing is providing a closed model that's only available online through an API so yeah that's the real story it's a despite the the name which is a misnomer in my mind yeah yeah do you see people users uh in your user base do you see them converging on a specific LLM provider for for the generative aspect or do you guys handle that yourself with the with the core product offering it depends on the customer really like so we have customers who are using you know Google Cloud Platform as their primary yeah cloud or AWS Azure so they tend to wanna start with that infrastructure and then access the LLM through that cloud providers private connections so that's something that's on our roadmap to provide for our enterprise customers um it's like if you download Plotly Studio now which I encourage everyone to do it's free you can use it I'm gonna check it out after this I got 100% needing some data visualization yeah you can do it on Mac it's a desktop app right you can do it on Mac and Windows but when you when you fire it up and you load a dataset when you hit that first button it's making a call so to the public LLM but in that's that's fine for like individual uses but enterprises typically want to go through the security path of their right yeah provider so like with AWS we're going to provide that through access to the LLM through AWS Bedrock for instance and we haven't encountered very many customers yet who are running their own LMS locally we see a lot of individuals doing it um testing maybe using it as part of the workflow to minimize their token consumption and cost but but for production apps they're using you know an API call to one of these LMS mmm hmm say the problem is it's just production apis are just so much more reliable than yeah and they have the latest version will be uh yeah the latest versions of the models right so we do think that that's gonna change I mean of course the cost should come down yeah but also it's it's not coming down really fast enough to make up for the huge increase in use so we're expecting more people probably to start reducing their cost through running local models because you know the the hardware uh depending on what you're trying to do for the use case you could have the hardware available and just yeah you know at least defray and and segment off those use cases to to lower your token usage interesting question are there like longer running processes with with a plotly that that would be better suited to like a you know a local LLM where it's like you know you turn it on you leave it on overnight and the next morning you come in and you have have what you're looking for or is plotly all focused around like really fast iteration times um our user experience is around fast iteration times but I would say like internally I mentioned before that you know every night we're testing hundreds of data sets and and generations and our internal AI engineering infrastructure is evolving pretty quickly to do this and and that's sort of a whole new you know I think our engineering team has developed like how do we what is the Agenic SDLC for us right that's a whole another topic that you should have our engineering leader Ben possibly to tell you about which supers and hey yeah hook us up um but you know they basically invented what it looks like inside a plotly for that interesting okay but I would say like I think that internally for us when we think about the cost of our internal stack and our operational profile that is where we're we're looking for cost efficiencies so where we could use local LMS if if they're good enough and give us the right the best result that's the thing like the product the end product experience can't suffer from these right yeah kinds of internal cost efficiencies that our engineering team would do so but that's the next thing we'd like to look for is is that kind of thing but on the user experience side you know right now we're doing basically this data to insight data to app in one shot experience yeah you know there's a lot of other functionality we're thinking about exploring there into data science workflows those could be more long running right yeah but for now the experience is more interactive and instantaneous in the development phase and then you publish the app and that app's on Plotly Cloud and you allow your users to interact with that app interesting got it I see very cool well as we're starting to wrap up I one question I had that I really wanted to ask you because I think you have a unique perspective where you've been in tech SAS now you've been part of this company that is kind of tech SAS moving into AI which I think a lot of companies are trying to do exactly what you're doing a lot of them are not as successful sounds like as plotly has been with Plotly Studio so what advice would you give to other companies other startups that are trying to build and transition in this to maybe more of an AI based platform build an AI tool similar to what you have done with plotly that's a good question I think as far as what you're delivering focus on what is the outcome for your audience and how come AI yeah if you if it's a tool if it's a software product right for other users think clearly around what type of users to solve what kind of problem what kind of outcome and then secondly think about how AI could help improve augment accelerate that for the user it's not about a parlor trick of what AI can do like everybody can sort of is you know has been experimenting with that over the last two years and there's a lot of entertaining parlor tricks that you can do with chat GPT sure but it's like why does it matter who cares like are you really helping solve a problem is it more than just a quick demo can I really use this to change my business and I I do think that's the phase that we're in now is yeah more AI sophistication yeah business decisions people are looking and we and we've all seen the reports around how many projects fail how there's a new the MIT study that has been circulating in the last week or so yeah around what was the stat something like 95 percent of Gen AI projects fail yeah right I mean a lot of that reason is because you know it's it's like oh I have this tool what can I do with it which is backwards you have to first think about what am I solving for the user and what can I solve now in a better faster way maybe that I couldn't before because I have access to AI so that's yeah and that's my advice focus on the outcome secondly how AI can help you do that and be a disruptor in the space yeah and thirdly think of like what's the audience how will you attract this audience if you don't already have it audience matters and you can't be everything to everyone and this is my marketing advice right like you really have to think about who your audience is and mm hmm if you're a young early company new company get super narrow like if you're building the product like just start serving a handful of people yeah and and be super close to them as you iterate um and let you'll be surprised too at the feedback you get and will you'll have new ideas about really what's the potential audience you could serve it was not in our crosshairs really to create a tool for lay people or right the ordinary data analysts but as we started to build the next iteration of plotly dash enterprise we were iterating on the development environment like how do people create the apps in the first place they can do it with plotly dash open source just from a command line but how can we make this a whole lot easier and it was in pulling the thread to make that easier and easier yeah that we thought well should it be a gooey where you click and drag that's sort of what the path we had been on yeah and then that was abandoned once we realized wait let's try it in AI native way if you just threw out all of that all of that and like what is an AI native way to to solve the same problem we don't know what that looks like but was in that exploration that spawned this new product Poly Studio I love it awesome people are turning it on its head you gotta you gotta start with yeah I think that there is I I hear and see this a lot from from new startups where they're talking about like oh you know our total addressable market is 2 billion people or whatever and they're focused on like having the biggest numbers their total accessible markets like no no no like you know I'm I can't speak too much I'm not like a successful startup founder yet um but people just overlook the fact that like they need to get a a slice of the pie it doesn't matter how big the slice of the pie is to start but like you gotta get that slice yeah yeah yeah and I did I and I should reconcile that now that I've thrown out that 1 billion knowledge workers number like that is our our the aspirational audience we want to speak to but that's not how we execute go to market quarter to quarter right yeah it is a subset and of that line to where for us what makes sense of for us as a business of where our audience already is and how can we go faster bigger in those areas so without revealing too much about that play to your strengths right yeah yeah and uh but it is a new North Star now that we see what's possible yeah with these kind of tools that are AI powered it's exciting that's a lot of a lot of exciting stuff I'm excited to get my vibe my vibe analysis on after this yeah I've been needing something like this for stuff I'm doing at work so I had it I had it in my to do list for today was go through the big panel a bunch of other stuff I think this is what we got to check it out not just saying that because you're on the podcast with us yeah well we're like sort of we're doing an event on September 23rd called The Rise of Vibe Analytics so that to basically talk about like what is the state of the art and what first of all what does this mean vibe analytics vibe analysis what is the experience but also like sharing from experts that we've come to know through our early access program like what innovations are they doing by using this approach in their ordinary work so we we mean for it to be an educational session as much as an unveiling of what you can do with with Plotly Studio for this well uh what's the link is there is there like a a link for that event already where's that yeah it's already up I'll yeah I'll share the link for you okay sweet yeah great just say we'll we'll put that in the in the yeah video description and everything yeah similar vein Dominic if people want to follow you follow plotly what are the best channels best ways for people to do that yeah you'll find plotlyat plotly.com um you'll you know we're on LinkedIn and all of the socials uh for me you can find me on LinkedIn just slash Dominic OK d O m E N I C um that's easy it's pretty easy we're on blue sky we're on Twitter OK x as they call it you're right yeah and we have a we have a great YouTube channel with with lots of videos like describing the how to's and we're starting to create more on flatly studios that comes out as well okay awesome excellent well Dominic we're excited to see what happens in the coming months as as especially as studio progresses we'll have to yeah we'll have you on again in you know three six months something like that and see yeah what it's turned into cause I'm sure it's gonna be different than what it is now the way things are going it is yeah I I love that I really appreciate the time