Bridging Data Science and Generative AI
Episode Summary
The two halves of a data organization are converging. Domenic Ravita, VP of marketing at Plotly, describes generative AI teams starting to join the classical machine learning and data science teams they used to sit apart from, because both need the same data quality and the same preparation. What that convergence produces is not another reporting tool but a different kind of software: custom data applications, written in Python, that put machine learning, generative AI and a real interface in one place. His examples are a market data product sold to customers and a clinical trial operation that cut the time of the trial itself in half. His argument is that you cannot buy either one. There is no app off the shelf that does this, and the internal tools that promise to build it fall short, so the data science team that understands the problem has to build the application itself.
Key takeaways
His opening observation is organizational rather than technical, and it is the whole reason this conversation exists. The teams running generative AI experiments, which often sat as a separate group on another side of the company, are starting to join or work more closely with the traditional data science teams doing classical machine learning
The reason for the merger is unglamorous, and it is the part an executive can act on. Both groups rely on the same kinds of data quality and data preparation, so they want to share that infrastructure
He credits data scientists with their skepticism of generative AI rather than treating it as resistance. The skepticism is good, he says, and data scientists are in the job of reliable answers, or at least of being able to tell you how reliable an answer is; it can make them somewhat slower adopters for their mainline work in some cases, depending on the use case
Where generative AI is actually safe in a data science workflow is narrower than the marketing suggests. The use cases are usually the less risky ones, and can start in a developer’s own environment: data shaping, looking something up that would have meant going to documentation, searching your own GitHub repository
He draws the line at the output rather than at the tool. Generative AI is not relied on for a final answer to a customer or a final interpretation of an analysis, but it can be used in the workflow of creating one, and upstream of that in shaping the data
His history of analytics has a turning point, and it is not a product. The last decade of open data science is where the macro multi-decade trend of open source came fully home to roost, with most development happening in open source analytics code, mostly Python
He does not assert Python’s dominance; he points at the scoreboard. Every survey he names, the TIOBE survey and the Stack Overflow survey, points to Python as the leading language, growing beyond data science
The term he uses is bigger than the one the industry uses. Some people call this BI as code, but he says it is larger than BI itself: data analytics as code, which is a distinct change from a monolithic interface-first visual analytics tool
His one-line description of the product is the clearest thing in the episode. Plotly Dash is Tableau turned inside out and made accessible via Python
The convergence argument has a mechanical basis he states plainly. Everything being done in AI is done in Python, or mostly Python, so the AI ecosystem and the analytics as code ecosystem have come together
He extends the argument past the data team, and credits the idea to Apple. Everyone can code is mainly aimed at kindergarten through college, but he thinks any career will have some element of coding in it, because that is what lets you work with LLM-generated code
What the approach buys an executive is ownership, stated as three things rather than one. You can own the code, own the data, and own the AI capability you are realizing inside that function
The deployment reality behind that is specific and it is not a preference. Most of Plotly’s customers run it on-premise because they need it air-gapped or because it is a regulatory situation, so these are already sensitive data environments
The internal example is a pharma clinical trial application, and the business case is time rather than insight. A pharma customer built a custom data application for clinical data science that automates, orchestrates and analyzes the whole trial process, and the outcome he gives is cutting the time of the trial itself in half
He makes the central claim twice, and it is the one the episode turns on. There is not really an app off the shelf that can do this at scale effectively, so you have to build an app like this for yourself
The customer-facing example is a data product sold to customers rather than an internal tool. S&P Global’s Capital IQ, part of its market intelligence offering, is a Plotly Dash application that lets prospects and customers see what is in the data, explore it and run analysis on it, with generative AI summaries built in to help describe it
What the clinical trial application actually does is coordination, and he says the data science team supporting the trial is the one positioned to build it. It covers activation and the enrollment forecast, external benchmarks, site selection, patient selection, scenario planning and simulation of how the study is progressing, with metrics and KPIs across therapeutic areas
The outcome he gives for it is a decision about the company’s own boundary. Building that application allowed the company to basically insource the clinical research organization, which he says they spoke about at an open data science conference event
On agents he asks for a definition before he answers, which is unusual for a vendor. A lot of what is being called an agent today reminds him of heuristic-based automation, and he says there is nothing wrong with that, because AI-driven automation is where much of the productivity is
His verdict on autonomous agents is scoped to the stakes rather than to the technology. For a decision like the expensive clinical trial process he has just described, an agent that independently makes the decision is clearly not ready for those kinds of high-impact decisions
The reason he gives is a property of the systems, not a stage of maturity. Agents are non-deterministic, so where a deterministic result is needed the simpler and cheaper answer is an imperative tool or a statistical or classical machine learning score, rather than something that could be a hallucination
He does put agents in a sequence, which keeps the door open. AI-driven automation on open Python frameworks is the foundation on which a reliable agent-based workflow will be built in future, and in the meantime there is a lot of low-hanging fruit that might sound boring
The production number he cites is his recollection of a range, and it is why his company exists. Statistics or studies from a few different organizations put somewhere between 80 and 90 percent of data science projects as never reaching production, with different companies solving different blockers
His account of what a data application is becoming is a description of his own product’s direction. It is no longer only a web application with interactive visualization and read and write controls; it now incorporates generative AI and is starting to look like a compound AI system built from an open ecosystem of building blocks
He puts a name on why that has to be modular. A composable application built from building blocks is not something you can do with a monolith or a traditional interface-driven visual analytics product
His forecast for the enterprise is standardization rather than more experimentation, and he gives the reason. It feels a little bit like the Wild West in how companies are experimenting with generative AI, and the expectation at his company is that it becomes a little more operationalized and standardized, because it has to be for security reasons
The analogy that carries the second half of the episode is the one Christina singles out on air. The biggest general purpose LLMs are starting to become a kind of commodity, and he says a lot of the people implementing them now see them as akin to databases: you do not use one database type for every use case, you describe your workload and pick the one with the right price performance fit
The consequence he draws from it is a build-or-buy decision framed around competence. What you use depends on your business intent and where your core competencies are, which is what decides between something off the shelf from an external provider and running a model like a Llama 3.1 in-house
He is careful about which part of in-house is affordable, and the distinction matters. Running a model in-house is becoming cost-effective and retraining a large one would be expensive, but small models can be cost-effective to retrain, and in any case you can augment with RAG or with contextual documents in the context window
Asked for resources he gives two, and points past the home page to what is on it. plotly.com carries example applications in a gallery and user stories across the industries the company works in, and the second resource he gives is his own LinkedIn profile
He is explicit about where his company stops, which is rarer than it sounds. Plotly is fairly unopinionated about what you do upstream, the data engineering tool chain and the model development, including whatever MLOps technology you use to prepare the model
The problem he says the product exists to solve is a staffing gap, not a technology gap. Data science teams are not typically full-stack software developers, so building the application looks daunting, and the two alternatives are outsourcing it to another department or using in-house visual analytics tools that fall short
His closing advice to executives names the place to look rather than the thing to buy. Look inside the business for opportunities to create operational efficiencies and differentiating capabilities through custom data applications, for internal use and for customers, in the lab, the clinical trial and the field
The reason he gives for that advice is the same claim twice, and it is the note he ends the argument on. You are not going to find apps off the shelf to do these, and you are not going to find internal tools that help you build them
About Domenic Ravita
Domenic Ravita is VP of marketing at Plotly, the company behind Plotly visualization and Plotly Dash, the Python framework data science teams use to build interactive data applications. Years before, he worked as a field engineer with customers on real-time event data, merging it into actionable analytics in what he describes as monolithic, interface-driven visual analytics tools. His argument on this episode is that data science and generative AI have converged on the same Python ecosystem, and that the applications this makes possible, from a customer-facing market data product to an AI-driven clinical trial operation, are not available off the shelf and have to be built by the teams that understand the data. This is his first appearance on AI Realized.
In this episode
| 00:58 | Welcome, and who Domenic Ravita is |
| 01:28 | The last time they spoke |
| 01:40 | The opening question: is this an exciting time for data scientists? |
| 01:54 | Why data scientists are attuned to what is happening |
| 02:18 | Generative AI teams joining the classical machine learning teams |
| 02:38 | The shared reason: the same data quality and the same preparation |
| 03:06 | Slower adoption for mainline work, and the Plotly community |
| 03:28 | Trusted results from probabilistic models |
| 03:39 | Starting with the less risky use cases |
| 03:58 | Data shaping, documentation, and searching your own repository |
| 04:24 | A thought partner, not the final answer |
| 04:50 | Upstream in the workflow, and where visualization comes in |
| 05:05 | How data and AI come together in code-based analytics |
| 05:18 | Field engineering, real-time event data, and monolithic tools |
| 05:51 | Tableau looked like magic, and then open data science happened |
| 06:31 | TIOBE and Stack Overflow, and Python as the leading language |
| 06:58 | BI as code, analytics as code, and Tableau turned inside out |
| 07:27 | Two ecosystems converging on Python |
| 08:05 | Everyone can code, and what it means for any career |
| 08:40 | A step function for operational decision-making |
| 08:50 | Owning the code, the data and the AI capability |
| 08:59 | On-premise, air-gapped, and regulated by default |
| 09:29 | Modular, open and extensible, and why that suits data assets |
| 10:13 | Examples from regulated environments |
| 10:37 | S&P Global, and what a market intelligence data product needs |
| 11:08 | Capital IQ as a Plotly Dash application |
| 12:14 | What a clinical trial costs in people, data and coordination |
| 12:44 | The custom application that cut trial time in half |
| 13:46 | No app off the shelf can do this at scale |
| 14:24 | Why the data science team is the one that has to build it |
| 14:39 | The question asked again, in different words |
| 14:51 | The same two examples, at greater length |
| 15:17 | Capital IQ, the data product sold directly to customers |
| 16:31 | Multiple datasets, and more than one way to manage the application |
| 16:58 | Pharmaceutical development, and customers like Amgen |
| 17:24 | Coordinating people, processes, data and analysis |
| 18:13 | Every step: enrollment forecasts, site selection, simulation |
| 18:57 | Earlier decisions, and insourcing the clinical research organization |
| 19:20 | The conference talk, and why he is bullish on this market |
| 20:01 | A case for agents? |
| 20:04 | What people mean by an agent, and heuristic-based automation |
| 20:48 | Not ready for high-impact decisions |
| 21:16 | Non-deterministic, and the cheaper deterministic answer |
| 21:58 | Low-hanging fruit, and the foundation agents will be built on |
| 22:45 | From pilots to production |
| 23:12 | 80 to 90 percent of data science projects never get there |
| 23:36 | The application as a compound AI system |
| 24:01 | Composable applications, and what a monolith cannot do |
| 24:48 | Where AI is going for the enterprise |
| 25:28 | From the Wild West to operationalized and standardized |
| 25:58 | Local development, local AI, and LLMs as a commodity |
| 26:29 | Picking a model the way you pick a database |
| 27:01 | Specialized models, and models not built on language at all |
| 27:45 | What is cost-effective to run, retrain or augment in-house |
| 28:11 | Own your own intelligence, and keeping precious data inside |
| 28:55 | The same analogy applied to cloud and on-premise |
| 29:23 | Resources |
| 29:31 | plotly.com, the gallery and the user stories |
| 30:07 | Resources for choosing a model and designing the application |
| 30:18 | More sophisticated Python and gen AI components, and the focus on production |
| 30:38 | The last mile, and what comes together in the application itself |
| 31:46 | Data science teams are not full-stack developers |
| 32:10 | Outsourcing, BI tools, and why both fall short |
| 32:49 | What listeners should take away |
| 33:27 | Look for the operational opportunities inside the business |
| 33:54 | You will not find these apps, or the tools to build them |
| 34:38 | Close |
In Domenic’s words
“Data scientists are in the job of reliable answers, or at least they can tell you how reliable an answer is.”
Domenic Ravita (02:38)
“the way I like to describe Plotly Dash is if you took Tableau and turned it inside out and made it all accessible via Python”
Domenic Ravita (06:58)
“they can cut the time in half of the trial itself”
Domenic Ravita (12:44)
“there’s not really an app off the shelf that can do this at scale effectively”
Domenic Ravita (13:46)
“it’s clearly not ready for those kinds of high impact decisions”
Domenic Ravita (20:48)
“anywhere from 80 to 90% of data science projects don’t make it to production for various reasons”
Domenic Ravita (23:12)
“It’s not one super LLM like a GPT-4 from OpenAI.”
Domenic Ravita (26:29)
“You’re not gonna find internal tools that help you build that.”
Domenic Ravita (33:54)
Resources
Domenic Ravita
Domenic Ravita on LinkedIn: The profile he sends listeners to on air, telling them to look for Domenic. He also points people to open data science events, Plotly events and Python events
Plotly: The company he markets for, and the address he gives on air. The site carries the gallery of example applications he mentions and user stories across financial services, pharmaceuticals, energy and manufacturing
Ideas and terms discussed
Analytics as code: His term for the shift, and he says it deliberately over the narrower BI as code, because what gets written in code is larger than business intelligence. It replaces the monolithic interface-first tool with something assembled from open source parts
Plotly Dash as Tableau turned inside out: The one-line description of the product. The point of the metaphor is access: everything the tool does is reachable from Python rather than from a fixed interface
Open data science: His name for the last decade in this field, which he describes as the macro multi-decade trend of open source coming fully home to roost in data science
Everyone can code: The idea he attributes to Apple, aimed mainly at kindergarten through college. His extension of it is that any career will have some element of coding in it, because that is what lets a person work with code an LLM generates
A custom data application: The thing the whole episode is about. Not a monolithic visual analytics tool and not a bought product: software built by a data science team for one problem, bringing a model, API calls, datasets and an interface people can act in together in one place
The last mile: Where he says his company operates, and where it stops. Upstream data engineering and model development are somebody else’s tools; the last mile is deploying, managing and securing the application and making it take actions
A compound AI system: What he says a data application is becoming, built from an open ecosystem of building blocks, which is why he also calls it composable. His argument is that a monolith cannot be assembled this way
Heuristic-based automation: His reading of much of what is called an agent today, offered without complaint: a lot of it reminds him of heuristic-based automation, and a great deal of productivity comes from automation, including AI-driven automation
Non-deterministic: The property that decides where he will and will not use an agent. When a deterministic answer is needed, an imperative tool or a classical machine learning score is simpler, cheaper and not a source of hallucination
LLMs as databases: The analogy Christina singles out on air. You do not use one database type for every workload; you describe the workload and pick for price and performance, and he expects model selection to work the same way
Local AI: What he says is coming, and what he talks about is cost. Running a model like a Llama 3.1 in-house is becoming cost-effective, small models can be retrained affordably, and RAG or contextual documents can cover the rest
Named on air
S&P Global and Capital IQ: His first customer example, named directly. Capital IQ is part of S&P Global’s market intelligence offering, and he says it is a Plotly Dash application, customer-facing, with generative AI summaries of datasets built in
Amgen: Named as an example of the customers Plotly has in pharmaceutical development. He does not attribute the clinical trial application he describes to any named company
UK Power Networks: Named once, at the very end, as a customer where the opportunity is operational decision-making in the field
Tableau: His own reference point, used twice: first for how magical monolithic visual analytics looked years earlier, then as the thing Plotly Dash turns inside out
TIOBE and Stack Overflow: The two surveys he cites for Python being the leading language and growing beyond data science
GPT-4, OpenAI and Llama 3.1: The models he names to make the database analogy concrete: not one super LLM, and a model you can bring in-house
GAI Insights and own your own intelligence: Raised by Christina Ellwood rather than by Domenic Ravita. GAI Insights is Paul Baier’s firm, and its model of own your own intelligence is the idea that a company should keep and build on its own data
Related AI Realized episodes and events
Analytics as Code: Why AI Stops Guessing With Data: Chris Parmer on what putting an analysis in code changes about what AI can be trusted to do with data. Parmer co-founded Plotly, the company behind Dash, the framework Domenic Ravita describes as Tableau turned inside out.
The Data Itself Is the First Prompt in Vibe Analytics: Domenic Ravita’s second appearance on AI Realized, less than six months later, on vibe analytics and what changes when the data rather than the prompt sets the terms.
Frequently Asked Questions
-
Analytics as code is building analysis, visualization and the application around them in code, usually Python, instead of inside a monolithic interface-driven tool. Domenic Ravita of Plotly says some people call it BI as code, and that what is being written in code is larger than business intelligence itself. His compressed description of the resulting product is Tableau turned inside out and made accessible via Python. What makes it possible is that most data science development has happened in open source analytics code for a decade, and the surveys he names, TIOBE and Stack Overflow, now put Python at the top.
Transcript 05:51 to 07:27
-
Most data science projects never reach production, and the range Domenic Ravita of Plotly gives is 80 to 90 percent of them, for various reasons that different companies are solving separately. He is careful that this is his recollection of statistics or studies done by a few different organizations rather than a citation. The one blocker he names is the application layer: what has to be produced is no longer just a web page with charts but something incorporating generative AI, closer to a compound AI system, and building that from an open ecosystem of parts is what gives more flexibility and more options for building just what is needed.
Transcript 22:45 to 24:01
-
Generative AI fits into the mundane parts of the workflow first, not the final answer. Domenic Ravita of Plotly puts the safe uses in a developer’s own environment: data shaping, looking up something that would otherwise mean a trip to documentation, searching your own GitHub repository, and acting as a thought partner. The line he draws is at the output rather than the tool. Generative AI is not relied on for a final answer to a customer or a final interpretation of an analysis, but it can be used in creating that analysis and upstream of it in shaping the data.
Transcript 03:28 to 05:05
-
AI agents are not ready to make high-stakes operational decisions on their own, and Domenic Ravita of Plotly is specific about why. Agents are non-deterministic, so for a decision like running an expensive clinical trial, an autonomous agent that independently makes the call is clearly not ready. Where a deterministic result is needed, the simpler and cheaper answer is an imperative tool or a classical machine learning score, rather than something that could be a hallucination. He also asks what people mean by an agent, since a lot of what carries the label reminds him of heuristic-based automation, which he has no objection to.
Transcript 20:01 to 22:45
-
AI is used in clinical trial operations to coordinate and analyze the whole process rather than to run any single step, and the application is custom-built. Domenic Ravita of Plotly describes a pharma customer whose data science team built a data application that automates, orchestrates and analyzes the whole process: activation and the enrollment forecast, external benchmarks, site selection, patient selection, scenario planning and simulation, with metrics across therapeutic areas. The business outcomes he gives are cutting the time of the trial itself in half and, for that company, insourcing the clinical research organization.
Transcript 12:44 to 19:20
-
A company should choose an LLM the way it chooses a database, by describing the workload and finding the right price performance fit. Domenic Ravita of Plotly makes the analogy directly: the biggest general purpose models are starting to become a kind of commodity, and there is no more a single model for every use case than there is a single database type. What decides the rest is business intent and where the core competencies are, which is the choice between something off the shelf from an external provider and running a model like a Llama 3.1 in-house. The same comparison covers cloud against on-premise: a price performance comparison first, then a security question about where the data sits.
Transcript 25:58 to 28:55
-
A custom data application is software a data science team builds for one problem, rather than a product bought to cover it. Domenic Ravita of Plotly describes one built for clinical trials that automates, orchestrates and analyzes the whole process, with the business outcome of cutting the time of the trial itself in half. His argument for building rather than buying is that there is not really an app off the shelf that can do this at scale effectively, and that the data science team supporting the work is the one that knows the most about the data and about which machine learning and generative AI should be applied to it.
Transcript 12:44 to 14:39
-
Data science teams build their own applications because outsourcing to a full-stack team and using a BI tool both fall short. Domenic Ravita of Plotly frames it as a choice a team faces once it can solve a problem with software: hand the project to another department that builds the full stack, or reach for the in-house visual analytics tools. Neither has the flexibility of an open Python-based framework, and data science teams are not typically full-stack software developers, so building it themselves looks daunting. What his company sells into that gap is the last mile: deploying, managing and securing the application.
Transcript 30:18 to 32:49
-
[00:58] Christina Ellwood: Welcome to AI Realized, a podcast for enterprise executives leading AI deployments. From tackling security data and operational challenges to navigating organizational transformation, AI deployment offers a unique opportunity to redesign organizations from the inside out. I’m Christina Ellwood, your host for today’s episode, and we’re talking today with Domenic Ravita, the vice president of marketing at Plotly. Welcome, Domenic.
[01:25] Domenic Ravita: Thanks, Christina. It’s a pleasure to be here.
[01:28] Christina Ellwood: We’re very happy to have you here, and I think we haven’t spoken to each other since AI Realized in October. Is that right?
[01:36] Domenic Ravita: Yeah. It’s... Time flies, right? That’s true. AI’s moving fast.
[01:40] Christina Ellwood: It certainly is. You’ve been a s- ardent student of data and AI markets for more than a decade. Are you excited about AI for the world of data scientists?
[01:54] Domenic Ravita: Yeah. It’s hard not to be. I think data scientists especially are attuned to what’s happening because more and more what traditional work they had been doing in machine learning, classical machine learning we might call it, often there was a separate group on another side of the company maybe who was experimenting with generative AI. And increasingly in- inside of corporate environments with our customers at Plotly, we’re finding that those experiments are starting to... Those teams doing generative AI experiments are starting to join or work more closely with traditional data science teams doing classical machine learning and data science. And the reason is because they’re really relying on the same kinds of data quality and data preparation, so they want to share that, that infrastructure. At the same time, they also are very skeptical of generative AI, which is good. Data scientists are in the job of reliable answers, or at least they can tell you how reliable an answer is. That’s a large part of it. And so depending on the use case, they may be a little bit more slower adopters in some cases for their mainline work, and we see this in even in talking with the Plotly community. But absolutely, everybody’s experimenting with it, and it’s a super exciting time in, in this area of data science and generative AI.
[03:28] Christina Ellwood: As you say, trust in the analytic results is vital in the data science world. So how are data scientists ensuring trusted results when they’re using probabilistic LLMs?
[03:39] Domenic Ravita: It’s not an easy thing to do. We find that it... Their use cases are usually less risky use cases, or it can start with just your own development environment so that you’re using generative AI to help you do things a little bit faster, a little bit more augmented. So we’re starting to see generative AI help with more of the mundane tasks that can be a part of a data scientist’s workflow, like just data shaping or things I would’ve taken a little bit more time to go look up elsewhere in documentation or maybe something in my own GitHub repository. It can work as a faster search for your own discovery. And we see that kind of use case with generative AI outside of the data scientist role in lots of different roles that it can be a good thought partner to help you. And so I think that’s one of the first ways that we see it being adopted in a way that’s, it’s not relying on generative AI for a final answer to give to a customer, for instance, or a final interpretation of a data analysis. But it can be used in the workflow of creating an analysis in even upstream of that in helping to shape the data. And for Plotly, part of that is Plotly visualization and Plotly interactive data applications.
[05:05] Christina Ellwood: So paint a picture for me, Domenic, about how data and AI are brought together in code-based analytics like Plotly offers.
[05:18] Domenic Ravita: Yeah. It’s, it-- The world has changed greatly in the last few years. Years ago, I was a field engineer working with customers, creating real-time event sort-- uh, real-time event data, and then trying to merge that together to create actionable analytics with visualizations and things like that. And the environments we were working in were more like a monolithic UI-driven visual analytics kind of technology, which was great. Years earlier, it looked like magic when you could do this kind of stuff with things like Tableau. But what’s been happening, I think, the last decade is this movement of open data science, where it’s a convergence of the macro multi-decade trend of open source technologies. In data science, it came fully home to roost, and most of that development has been occurring in open source analytics code It mostly is Python, although there have been other languages over the years like R and Julia, and then front-end languages like JavaScript. But every, every survey you look at now, like the TIO- TIOBE survey or Stack Overflow survey, all of them are pointing towards Python as the leading language, and growing even bigger beyond data science. So what’s that has... What’s that hasn’t... What it’s enabled is a larger ability to do analytics as code. Some people call this BI as code, but it’s, it’s larger than BI itself. It’s data analytics as code. And so that, that marks a distinct kind of change from using a monolithic UI first kind of visual analytics tool, because this gives you a lot more flexibility. It, and, uh, the way I like to describe Plotly Dash is if you took Tableau and turned it inside out and made it all accessible via Python. That is very much like the ethos of the kind of tool that... and the kind of flexibility it enables. And so that with AI, everything being done in AI, of course, is with Python, or mostly Python. So these two ecosystems of AI and analytics of code, and analytics as code have come together, and it’s this super rich opportunity to create all sorts of interesting data applications, analytics applications that incorporate classical machine learning along with what’s happening in generative AI. And there’s a bigger movement here, I think even beyond the analytics as code movement, which is this sort of everyone can code, which is an idea that Apple has put forth, and it’s mainly kindergarten through college. But I think it’s no matter what your career path is, there’s gonna be some element of coding that’s gonna be... That’s gonna make your... You have superpowers essentially, because you’re gonna be able to work more easily with ChatGPT or any LLM-generated code to augment what you’re doing.
[08:40] Christina Ellwood: It sounds like a big step function for operational decision-making by executives and mid-level operators. Is that right?
[08:50] Domenic Ravita: I think it is because it allows a lot more control over the code that you’re using, how you build. You can build more specific kinds of analytical and data applications more economically. You can own that code, you can own your data, you can own the AI or the AI capability that you’re realizing within that function. And just to give an example of that, we find a lot of our-- most of our customers are using Plotly in an on-premise environment because they need it air-gapped or it’s a regulatory situation. So these are already sensitive data environments. And so they really like the ability to have this modular approach where it’s open and extensible. And so I think that as every company is trying to develop their, the data assets within the enterprise and enrich their own data assets, and on top of that, they want to build their own AI capabilities that are unique and a core competency and a differentiator for their business. This approach of analytics as code and the approach that Plotly Dash has fits really well within that approach.
[10:13] Christina Ellwood: At AI Realized, we heard a lot about the applications that were being built inside regulated environments like banking and healthcare. Do you have some particular examples that you can use to illustrate how the combination of analytics as code and generative AI are being used together to generate value or ROI for companies?
[10:37] Domenic Ravita: Yeah. I, let me give you maybe two examples. Financial services is a really big vertical for us, and there’s a company, S&P Global, who provides lots of different products across industries. Market intelligence data is one of them. And so they were looking for a way to provide their own data products as more easily to their customers to also to describe what the data product is, but also to s- disseminate it and also demonstrate this. So they’ve built a data product, or which they call Capital IQ, and this is actually a Plotly Dash application. And so they’ve been able to... And this is the customer-facing application. So you can explore the S&P market intelligence datasets. You can start to do some analysis there. You can get the feed API, so it becomes a portal for customer engagement for them as well. And they’re incorporating generative AI in terms of summaries of datasets and things like that right into that Capital IQ product. So that speaks to the extensibility and flexibility of the kinds of data apps you can create with Plotly Dash Enterprise. A second case I think is really unique and quite powerful is what the pharma companies have done with Plotly, which is lots of different use case, but one in particular has created basically an AI-driven clinical trial planning process. So clinical trials for pharma companies require a lot of manpower. There’s a lot of data. They’re very expensive to run. And so there’s a lot of coordination that has to happen from the protocol design for the study, the study feasibility, site selection, and then of course the analysis of the data, tracking disease progression, and documenting and extracting your metadata, all of this. So we have a customer in pharma who’s built basically a custom data application for clinical data science that allows them to automate, orchestrate, and analyze the whole clinical trial process. So the business outcome for this is that they can cut the time in half of the trial itself Saving a pretty good financial sum in the process and have all of the people and the data coordinated in a way so that they can make data-driven operational decisions throughout the process from activation and enrollment and the forecast for that. Are they hitting benchmarks along the way? Do they-- have they optimized their sites? What about the patient diversity? What about scenario planning on that data if things don’t go as expected? Which that’s why you do the study, right? So you want to do these sort of what if study and analysis, what if scenarios, and study these simulations. So when I describe that, I think the listeners will see there’s not really an app off the shelf that can do this at scale effectively and incorporate machine learning, the clinical trial data, the people. And you have to build an app like this for yourself And in their case, a lot of these pharma company, it’s the data science team embe- who’s supporting the clinical trials who knows the most about the data and the various kinds of data science and machine learning and gen AI which should be applied to that. And so when the data science team is making that decision, they need something that suits them to build that kind of interactive data application. And that’s a situation where Plotly Dash Enterprise suits, suits that very well.
[14:39] Christina Ellwood: Domenic, how does the generative AI and analytics as code come together in a use case for a Plotly customer? Bring some color to that for us.
[14:51] Domenic Ravita: Yeah. There are lots of examples from financial services to energy and pharma, but maybe if I just start with two. S&P Global is a large financial services technology company, and they provide lots of data and solutions to companies across the financial services industry, as well as to other vertical industries. They have a product called Capital IQ, which is part of their market intelligence offering. So they’re providing unique market data assets to investors, hedge fund managers, marketers, and this is a set of data that they, they wanna share with their prospects and customers in a unique way that allows them to see the value of the dataset, explore it. They treat this as a data product, and that product is literally sold directly to customers. So they were looking for a way to disseminate that and share that product, but also for a way for prospects and customers to more easily see what’s possible and see what’s in the data and explore the data. So with Capital IQ, this is a customer-facing application built on Plotly Dash Enterprise that allows them to share that data, explore the data, do analysis on it, and they’ve also incorporated generative AI summaries and things like on that order to help for people to assist them in describing that data. And they’re not doing this just for one data set, but there’s multiple, so there’s multiple sorts of ways to h- manage the application internally, which makes this easier for them. So that’s one example of a external customer facing one. The second I’ll give you is an example of an internal application that helps with operational efficiency, and that’s for a use case we see pretty often in pharmaceutical development. We have customers in that area like, like Amgen for instance. And so in that area we see in like clinic- clinical trial operations is a great opportunity for this 'cause it’s, it’s very labor intensive in terms of staff of all levels from PhD scientists, virologists, oncologists, et cetera, to the clinicians, to people working on the trials. And so within clinical operations, there’s a large amount of work that has to be coordinated, the people, the processes, the data of course, and the analysis on that data. And the data science teams that support this clinical trial operation are really the ones in the best position to understand how to make that whole enterprise more efficient and streamlined. But they need a data driven automated way to coordinate the people and the data and the analysis and the machine learning and the models that are applied, and also the incorporation of Gen AI where it’s appropriate. And so they’ve used Plotly Dash Enterprise to build a custom data app that does just... That allows AI driven decision making and coordination throughout the clinical trial operation. So imagine every step of this process from the activation and the enrollment forecast to incorporating external benchmarks, optimizing site selection and patient selection, and then Doing scenario planning on all of this, and then simulating what the study may-- how the study is progressing throughout, throughout the whole process, and providing metrics and KPIs across therapeutic areas in the form of study metrics and KPIs. So this helps to get visibility from an operational standpoint so that better decisions can be made as early as possible in the process to ensure that this big, expensive operation of a clinical trial can go more easily. So that’s allowed this company to basically insource the clinical research organization. They spoke about this recently at an open data science conference event, and I think this is one of the really interesting use cases that within pharma, but in lots of different industries where the power of data and flexible applications that serve data science and AI, it opens up a huge amount of untapped potential for better in-the-moment operational decision-making. So I’m really quite bullish on this market for data and AI applications in areas like this.
[20:01] Christina Ellwood: Domenic, that sounds like a great case too for agents.
[20:04] Domenic Ravita: Yeah, I do think it depends on what kind of agent we’re talking about. I always want to start with understand what people mean by an agent. I think there are a lot of things being called agents today that remind me a lot of Just heuristic-based automation. And there’s nothing wrong with that. There’s a huge amount of productivity and operational efficiency to be gained by automation and even AI-driven automation. As you think about each step in a process, there can be an AI informed. It could be through classical machine learning, or it could be a rule based on the result of some score that’s been computed in the background. But the idea is right now, in terms of an autonomous agent that independently makes the decision for like the use case I just described, for a very expensive clinical trial process, it’s clearly not ready for those kinds of high impact decisions. That said, I’m, I think AI-driven agents are a very interesting area that has a lot of potential. But we have to-- But the way that they’re being applied now needs to be appropriate with their level of the kind of result they can give. They are non-deterministic. So in many cases, if you-- The simpler, cheaper answer is just to use the right kind of imperative tool or statistically or m- classical machine learning generated pro- probability result in terms of a score rather than something that could be a hallucination from an AI agent, for instance.
[21:51] Christina Ellwood: So we’re not quite ready for agents in that use case, even though they may sound like a good fit.
[21:58] Domenic Ravita: Yeah, I do think that there’s a lot of experimentation happening there. There’s... But there’s so much low-hanging fruit, too, in just... It might sound boring, it’s maybe not as sexy, but basically using the data that is available and the flexibility of this, of Python with open data frameworks like Plotly Dash to do that AI-driven automation. And I do think that is the foundation on which you s- will be able to, in the future, build a reliable agent-based workflow on top of that. But at the moment, if you, if you need, like, that deterministic result, that’s probably... You’re gonna use some sort of imperative or statistically based program for that.
[22:45] Christina Ellwood: And those two use cases that you just described are in production. Are you seeing a momentum in organizations going from pilots of generative AI empowered use cases to production?
[23:01] Domenic Ravita: We’re in the business at Plotly of getting data science applications into production. That is one sort of market level problem that we see. Uh, there have been statistics or studies done by a few different organizations in the last few years that give results anywhere from 80 to 90% of data science projects don’t make it to production for various reasons. And there are companies and people solving different problems as to why that is to remove those blockers. The part that we solve at Plotly is to... Is the application itself, which more and more these days is becoming not just a, a web application with interactive visualization and really sophisticated controls for a read/write application, but it’s also incorporating gen AI, and it’s starting to look more like compound, like a compound AI system. And it’s being built from the, this open ecosystem of building blocks. So it’s like a composable app, application in that way. And that’s, that’s only something that you can do If, you know, you have this sort of... You can’t do that with a monolith, with a UI-driven traditional visual analytics product. So that helps, I believe, provide more flexibility and more options for building just what you need. That helps to sort of the business case on get the motivation to get this into production, 'cause it meets the needs. You can see that. And you can really solve things that maybe, that you couldn’t solve before in s- in specific ways, specific to your business.
[24:48] Christina Ellwood: Okay. All right. Where do you see AI going in the coming years for the enterprise?
[24:58] Domenic Ravita: AI is everywhere right now. I th- we see it across individuals in helping to augment and accelerate tasks from everything from content creation and marketing to code generation in the engineering team, and lots of processes in between. It feels a little bit like the Wild West right now in the way that companies are experimenting with gen AI. Our expectation at Plotly is that it’s gonna become a little bit more operationalized and standardized within businesses b- because it has to be for security reasons. But also it needs to go that way so that we can, so that companies can really get some bigger advantage out of it and create durable advantages building on their data assets, creating new data assets as well, and creating their own AI capability. So we see a movement towards local development and local AI that helps to support that. LLMs are starting to become, at least the biggest of the general purpose LLMs, are starting to become a kind of commodity, and a lot of folks that are implementing it realize that there’s... It’s, they’re becoming akin to databases in this, in that there’s not a single database type that you use for all use cases. Rather, you describe your needs for a database workload and you find the appropriate database that has the right price performance fit for that. That’s the same thing that’s happening right now with LLMs. It’s not one super LLM like a GPT-4 from OpenAI. It has its place, but there’s many more types of language models that are specialized language models, or they’re not built on language, they’re built on chemical reactions and chemistry. We’re seeing this across the sciences and different sorts of neural network-based models like this being used to create other types of models that help predict. Essentially, they’re sequence, they’re token sequence predictors, and so there’s a wide variety of applications of that general idea. And so it’s very much what we see is that you’ll, based on what your business intent is and where you want your, where are your core competencies, that’ll help you decide what sort of LLM to use, something off the shelf from an external provider, or to bring that model like a Llama 3.1 and run it in-house. It’s becoming cost-effective to do that in-house. Or if you need to do some level of... You’re not gonna... It’d be expensive to retrain a model like that, but for small models, it is, it can be cost-effective to retrain them. But at least you can augment them in your enterprise with RAG or contextual docs put in the context window. Local AI, we think, is a big thing that’s coming essentially.
[28:11] Christina Ellwood: Yeah. I think that’s shared by the GAI Insights folks too in their model of own your own intelligence and some of the other startups that we’ve seen recently, like Napsack, that are providing ways for you to run a small language model even on your own laptop. So I think the idea that companies that have precious data would wanna be able to keep it inside and be able to fully leverage it w- without compromising it, I think is very logical, and the ability to make it robust and scalable for their operation. So I think that’s a good-- And I love your analogy to picking the database. I think that’s a really good one, that you pick the model and the type of AI you’re using based on your use case and, and scale it accordingly.
[28:55] Domenic Ravita: I think that model also fits the, the cloud scenario as to if you’re doing it in the cloud or on-prem. Uh, like the database, I think that’s similar to... It’s analogous to h- if you’ll have the LLM accessible through an API in the cloud or if you wanna run this on-prem, you st- you have a price performance comparison to do, you have a security concern to look at as to where the data is. And yeah, I think the ana-analogy fits pretty well.
[29:23] Christina Ellwood: What resources do you recommend for listeners who wanna learn more about that model or you or Plotly?
[29:31] Domenic Ravita: For Plotly, it’s pretty simple, plotly.com. We’ve got lots of example applications in a gallery that you can find there on the website, as well as user stories across, you know, the many industries we’re working in, financial services, pharmaceuticals, energy, manufacturing. And for me, it’s pretty easy. You just go to my LinkedIn profile, look for Domenic. I think I was one of the early ones on LinkedIn. Yeah, and you’ll s- you’ll find us at open data science events, Plotly events, Python events, excuse me, and basically on the data science scene.
[30:07] Christina Ellwood: What about resources for the process of selecting your model and designing your application? You have any thoughts there? Resources?
[30:18] Domenic Ravita: Yeah, we, yeah, we, we’re primarily focused on getting these data applications into production. And they’re-- these applications, like I said, are becoming more sophisticated in terms of the third-party Python and machine learning and gen AI components that they incorporate. So we, we basically solve that last mile gap to make this application actionable and literally take actions in the application. But we’re fairly unopinionated about what you do upstream in terms of preparing your data, the data engineering tool chain that you’re using, or in your model development. So you may use some sort of MLOps technology to help you prepare the model. We’re not really en- engaged in helping do those upstream tasks, but once you have your model and you understand the users as a data science team that you need to serve, and you wanna bring together the model, any API calls, the datasets, either embedded or that you’re gonna access remotely, all of that comes together in the production data application that we help you deploy, manage, secure And also help to match your unique look and feel in terms of user experience for your audience. And that’s something that data science teams, when they’re n- when they find that they’re in a position to solve a problem through some sort of custom data application, they’re faced with a choice. They’re not full stack software developers typically, so that looks daunting, and they could outsource the project to another department in their company to go build the full stack. Or they could try to use some in-house tools themselves, maybe these visual analytics tools like a BI tool. But both of those situations or choices fall f- pretty short, and that’s because they don’t have the flexibility. They’re not b- open Python-based frameworks. And Plotly fills that middle space to help data science teams in particular conceive of, dream big, and get their AI-driven and machine learning-driven data applications into production inside of their corporate environments, or in, like with S&P Global, share them publicly with customers.
[32:49] Christina Ellwood: That’s great. I love that example of you being involved in the last mile. I ha- think that helps listeners to imagine where they would begin to look to leverage Plotly’s capability. As we wrap up, what would you like our listeners to take away from our conversation today?
[33:06] Domenic Ravita: I would say this is a moment in time where y- there’s more and more data accessible to you as a knowledge worker, but also a data scientist, and there’s more and more compute and AI-driven capabilities from this open data science ecosystem. So what I would encourage listeners to do, and for executives, is to, within their businesses, to look for these opportunities to create operational efficiencies, create innovation, create new differentiating capabilities for their business through these unique cr- custom data apps that they could create across their enterprise internally, but also for customers. And you’re not gonna find apps f- to do these off the shelf. You’re not gonna find internal tools that help you build that. So that’s what I would encourage them to do, 'cause there’s so much to be So much potential for data and AI for operationalizing better decision-making in the field, in the business operations re- realm, in the lab, in the clinical trial, in the field, for instance, like our customer UK Power Networks. Look for those operational, those op- opportunities for better operational decision-making with data and AI, and look at data applications with, with Python as a way to do that fast.
[34:38] Christina Ellwood: Okay. Thank you so much, Domenic Ravita, VP of marketing at Plotly. I really appreciate you joining me today on AI Realized.
[34:46] Domenic Ravita: Thanks so much for having me, Christina. It’s been a pleasure.