StarZero All shows Try the podcast clipper →

Mar 20, 2024 · 26m · mad

2024 will be the year of ENTERPRISE AI | Florian Douetteau, CEO of Dataiku

Florian Douetteau · 19m spoken Matt Turck · 4m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

Dataiku CEO Florian Douetteau joins Matt Turck on The MAD Podcast to discuss the evolution of enterprise AI, explaining how companies are moving from experimental generative AI projects to production-ready, hybrid architectures. Douetteau outlines Dataiku's end-to-end platform philosophy, the launch of the LLM Mesh, and practical strategies for cost governance, model agnosticism, and AI democratization.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 18.7% of the talking time here. How this is scored →

Matt as informed peer 3.6 Guest teaching 4.1 Guest disagreement 1.0 Matt pushing back 2.3
05100:0010:0020:001:09–4:33 · Matt as informed peer 2/10 Dataiku's Enterprise Scope and Core Mission Matt opens with a broad industry question regarding enterprise readiness for AI in 2024. Florian clarifies terminology by distinguishing between mature predictive/tabular ML and early-stage generative AI experiments.4:33–8:50 · Matt as informed peer 3/10 Concrete Use Cases: Tabular Machine Learning vs. Generative Personalization Matt prompts Florian to break down the differences between tabular data AI and generative AI. Florian details how generative AI acts as a last mile layer of personalization on top of traditional recommender systems.8:50–12:26 · Matt as informed peer 4/10 Strategic Advice for Enterprise Generative AI Deployment Matt steers the conversation to enterprise concerns around ROI and wasted expenditure. Florian offers candid insight into API vendor switching and shares an internal example where Dataiku almost wasted $100k on an internal AI app before adding cost controls.12:26–14:31 · Matt as informed peer 3/10 Adoption Blockers: Security, Guardrails, and the Empirical Phase of GenAI Matt lists potential adoption blockers like skills and governance. Florian reframes current GenAI development as empirical dark science or cooking a sauce, where teams tweak parameters until it works without full predictability.14:31–16:34 · Matt as informed peer 2/10 Dataiku Product Architecture and Non-Coder Accessibility Matt asks for a product tour of Dataiku's platform architecture. Florian explains their vision of unifying the full data lifecycle while self-deprecatingly admitting his own coding skills have aged to highlight the platform's accessibility for non-coders.16:34–21:10 · Matt as informed peer 5/10 Product Philosophy: Platform vs. Point Solution Matt asks about product philosophy and later synthesizes the exact architecture of the LLM Mesh using Pinecone and multi-LLM routing. Florian details why building an end-to-end platform was a controversial but necessary bet.21:10–24:23 · Matt as informed peer 6/10 Hybrid AI Architectures and Data Pipelines Matt demonstrates high technical knowledge by asking whether modern AI architectures rely on hybrid setup chaining predictive ML output into generative layers. Florian validates Matt's hypothesis with examples from data pipelines.1:09–4:33 · Guest teaching 4/10 Dataiku's Enterprise Scope and Core Mission Matt opens with a broad industry question regarding enterprise readiness for AI in 2024. Florian clarifies terminology by distinguishing between mature predictive/tabular ML and early-stage generative AI experiments.4:33–8:50 · Guest teaching 4/10 Concrete Use Cases: Tabular Machine Learning vs. Generative Personalization Matt prompts Florian to break down the differences between tabular data AI and generative AI. Florian details how generative AI acts as a last mile layer of personalization on top of traditional recommender systems.8:50–12:26 · Guest teaching 5/10 Strategic Advice for Enterprise Generative AI Deployment Matt steers the conversation to enterprise concerns around ROI and wasted expenditure. Florian offers candid insight into API vendor switching and shares an internal example where Dataiku almost wasted $100k on an internal AI app before adding cost controls.12:26–14:31 · Guest teaching 5/10 Adoption Blockers: Security, Guardrails, and the Empirical Phase of GenAI Matt lists potential adoption blockers like skills and governance. Florian reframes current GenAI development as empirical dark science or cooking a sauce, where teams tweak parameters until it works without full predictability.14:31–16:34 · Guest teaching 3/10 Dataiku Product Architecture and Non-Coder Accessibility Matt asks for a product tour of Dataiku's platform architecture. Florian explains their vision of unifying the full data lifecycle while self-deprecatingly admitting his own coding skills have aged to highlight the platform's accessibility for non-coders.16:34–21:10 · Guest teaching 4/10 Product Philosophy: Platform vs. Point Solution Matt asks about product philosophy and later synthesizes the exact architecture of the LLM Mesh using Pinecone and multi-LLM routing. Florian details why building an end-to-end platform was a controversial but necessary bet.21:10–24:23 · Guest teaching 4/10 Hybrid AI Architectures and Data Pipelines Matt demonstrates high technical knowledge by asking whether modern AI architectures rely on hybrid setup chaining predictive ML output into generative layers. Florian validates Matt's hypothesis with examples from data pipelines.1:09–4:33 · Guest disagreement 1/10 Dataiku's Enterprise Scope and Core Mission Matt opens with a broad industry question regarding enterprise readiness for AI in 2024. Florian clarifies terminology by distinguishing between mature predictive/tabular ML and early-stage generative AI experiments.4:33–8:50 · Guest disagreement 1/10 Concrete Use Cases: Tabular Machine Learning vs. Generative Personalization Matt prompts Florian to break down the differences between tabular data AI and generative AI. Florian details how generative AI acts as a last mile layer of personalization on top of traditional recommender systems.8:50–12:26 · Guest disagreement 2/10 Strategic Advice for Enterprise Generative AI Deployment Matt steers the conversation to enterprise concerns around ROI and wasted expenditure. Florian offers candid insight into API vendor switching and shares an internal example where Dataiku almost wasted $100k on an internal AI app before adding cost controls.12:26–14:31 · Guest disagreement 1/10 Adoption Blockers: Security, Guardrails, and the Empirical Phase of GenAI Matt lists potential adoption blockers like skills and governance. Florian reframes current GenAI development as empirical dark science or cooking a sauce, where teams tweak parameters until it works without full predictability.14:31–16:34 · Guest disagreement 0/10 Dataiku Product Architecture and Non-Coder Accessibility Matt asks for a product tour of Dataiku's platform architecture. Florian explains their vision of unifying the full data lifecycle while self-deprecatingly admitting his own coding skills have aged to highlight the platform's accessibility for non-coders.16:34–21:10 · Guest disagreement 2/10 Product Philosophy: Platform vs. Point Solution Matt asks about product philosophy and later synthesizes the exact architecture of the LLM Mesh using Pinecone and multi-LLM routing. Florian details why building an end-to-end platform was a controversial but necessary bet.21:10–24:23 · Guest disagreement 0/10 Hybrid AI Architectures and Data Pipelines Matt demonstrates high technical knowledge by asking whether modern AI architectures rely on hybrid setup chaining predictive ML output into generative layers. Florian validates Matt's hypothesis with examples from data pipelines.1:09–4:33 · Matt pushing back 1/10 Dataiku's Enterprise Scope and Core Mission Matt opens with a broad industry question regarding enterprise readiness for AI in 2024. Florian clarifies terminology by distinguishing between mature predictive/tabular ML and early-stage generative AI experiments.4:33–8:50 · Matt pushing back 2/10 Concrete Use Cases: Tabular Machine Learning vs. Generative Personalization Matt prompts Florian to break down the differences between tabular data AI and generative AI. Florian details how generative AI acts as a last mile layer of personalization on top of traditional recommender systems.8:50–12:26 · Matt pushing back 3/10 Strategic Advice for Enterprise Generative AI Deployment Matt steers the conversation to enterprise concerns around ROI and wasted expenditure. Florian offers candid insight into API vendor switching and shares an internal example where Dataiku almost wasted $100k on an internal AI app before adding cost controls.12:26–14:31 · Matt pushing back 2/10 Adoption Blockers: Security, Guardrails, and the Empirical Phase of GenAI Matt lists potential adoption blockers like skills and governance. Florian reframes current GenAI development as empirical dark science or cooking a sauce, where teams tweak parameters until it works without full predictability.14:31–16:34 · Matt pushing back 1/10 Dataiku Product Architecture and Non-Coder Accessibility Matt asks for a product tour of Dataiku's platform architecture. Florian explains their vision of unifying the full data lifecycle while self-deprecatingly admitting his own coding skills have aged to highlight the platform's accessibility for non-coders.16:34–21:10 · Matt pushing back 4/10 Product Philosophy: Platform vs. Point Solution Matt asks about product philosophy and later synthesizes the exact architecture of the LLM Mesh using Pinecone and multi-LLM routing. Florian details why building an end-to-end platform was a controversial but necessary bet.21:10–24:23 · Matt pushing back 3/10 Hybrid AI Architectures and Data Pipelines Matt demonstrates high technical knowledge by asking whether modern AI architectures rely on hybrid setup chaining predictive ML output into generative layers. Florian validates Matt's hypothesis with examples from data pipelines.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 29.7% · guest 70.3%0:00 · Matt 29.7% · guest 70.3%3:00 · Matt 16.2% · guest 83.8%3:00 · Matt 16.2% · guest 83.8%6:00 · Matt 8.3% · guest 91.7%6:00 · Matt 8.3% · guest 91.7%9:00 · Matt 7.5% · guest 92.5%9:00 · Matt 7.5% · guest 92.5%12:00 · Matt 18.8% · guest 81.2%12:00 · Matt 18.8% · guest 81.2%15:00 · Matt 17.7% · guest 82.3%15:00 · Matt 17.7% · guest 82.3%18:00 · Matt 11.5% · guest 88.5%18:00 · Matt 11.5% · guest 88.5%21:00 · Matt 27.6% · guest 72.4%21:00 · Matt 27.6% · guest 72.4%24:00 · Matt 37% · guest 63%24:00 · Matt 37% · guest 63%
Sharpest disagreement ▶ 25:02 Playful teasing about VCs being tools

Florian turns Matt's startup advice question back on him by asking whether VCs are tools or platforms, prompting Matt to humorously acknowledge the double entendre.

 Hardest push from Matt ▶ 20:17 Synthesizing and validating the LLM Mesh model

Matt tests Florian's abstract description of the LLM Mesh by restating it as a concrete workflow involving vector databases like Pinecone and model governance layers.

 Biggest teaching moment ▶ 12:40 Explaining GenAI deployment as dark science and cooking

Florian educates Matt on the pragmatic reality of GenAI engineering, reframing it from clean software architecture to trial-and-error experimentation akin to cooking a sauce.

 Matt holds his own ▶ 21:11 Proposing hybrid model-chaining architecture

Matt displays deep domain expertise by asking whether enterprise systems connect predictive analytics engines directly into LLMs as input pipelines.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Dataiku's Enterprise Scope and Core Mission 2411 Matt opens with a broad industry question regarding enterprise readiness for AI in 2024. Florian clarifies terminology by distinguishing between mature predictive/tabular ML and early-stage generative AI experiments.
Concrete Use Cases: Tabular Machine Learning vs. Generative Personalization 3412 Matt prompts Florian to break down the differences between tabular data AI and generative AI. Florian details how generative AI acts as a last mile layer of personalization on top of traditional recommender systems.
Strategic Advice for Enterprise Generative AI Deployment 4523 Matt steers the conversation to enterprise concerns around ROI and wasted expenditure. Florian offers candid insight into API vendor switching and shares an internal example where Dataiku almost wasted $100k on an internal AI app before adding cost controls.
Adoption Blockers: Security, Guardrails, and the Empirical Phase of GenAI 3512 Matt lists potential adoption blockers like skills and governance. Florian reframes current GenAI development as empirical dark science or cooking a sauce, where teams tweak parameters until it works without full predictability.
Dataiku Product Architecture and Non-Coder Accessibility 2301 Matt asks for a product tour of Dataiku's platform architecture. Florian explains their vision of unifying the full data lifecycle while self-deprecatingly admitting his own coding skills have aged to highlight the platform's accessibility for non-coders.
Product Philosophy: Platform vs. Point Solution 5424 Matt asks about product philosophy and later synthesizes the exact architecture of the LLM Mesh using Pinecone and multi-LLM routing. Florian details why building an end-to-end platform was a controversial but necessary bet.
Hybrid AI Architectures and Data Pipelines 6403 Matt demonstrates high technical knowledge by asking whether modern AI architectures rely on hybrid setup chaining predictive ML output into generative layers. Florian validates Matt's hypothesis with examples from data pipelines.

Statements from this episode (10)

Assertion Not checkable as stated
Dataiku reached $230M ARR and 600 customers as of September 2023
“And back then, we crossed 600 customers. We were two thirty million of IRR. Back then, and we have more than 1000 employees.”
Florian Douetteau Mar 20, 2024 ▶ 1:37
Disclosure
Douetteau: Dataiku has about 100 customers using its platform for LLMs
“So we've got about 100 customers that used our platform for some LLM associated use cases.”
Florian Douetteau Mar 20, 2024 ▶ 4:00
Insight
Douetteau: Generative AI serves as the last-mile personalization layer for traditional AI
“Generative AI is providing this last mile, this additional benefit in terms of traditional AI, the last mile of, let's say, personalization, for instance, The last mile of accessibility in other contexts”
Florian Douetteau Mar 20, 2024 ▶ 8:26
Prediction Not checkable as stated
Douetteau: Successful AI apps will likely need to switch LLM providers
“Because of the evolution of technologies and LLM out there, it's very likely that you will have to switch from one provider to the other over the course of your application, if your application is successful.”
Florian Douetteau Mar 20, 2024 ▶ 9:17
Prediction Not checkable as stated
Douetteau: Enterprises will realistically run hundreds of GenAI apps in production
“It's actually fairly realistic for an enterprise to have hundreds Of Generative AI applications in production at some point.”
Florian Douetteau Mar 20, 2024 ▶ 10:10
Insight
Douetteau: Enterprises need a gateway between applications and LLMs for cost tracking
“You actually need to have a gateway between your application and your LLMs in order to track all of the costs and understand before moving things to production how much you can anticipate in terms of cost.”
Florian Douetteau Mar 20, 2024 ▶ 11:04
Insight
Dataiku CEO: Deploying enterprise GenAI is currently trial-and-error 'dark science'
“Because if we're, again, a bit honest, it's not that easy to make generative AI application work in practice. We are playing a little bit dark science there, where you get your LLM, your fine tune, your rag, your stuff around. And you decide, like, okay, good,…”
Florian Douetteau Mar 20, 2024 ▶ 12:51
Prediction Not checkable as stated
Dataiku CEO: Domain experts will not code anytime soon, despite AI Copilots
“And so I sympathize with, like, everyone that actually needs and have some domain expertise, but won't get into code anytime soon, even with Copilot or whatsoever, and that still need to leverage their data in order to get things done.”
Florian Douetteau Mar 20, 2024 ▶ 16:15
Assertion Not checkable as stated
Enterprises prototype with GPT-4 but shift to cheaper self-hosted models for production
“We see quite a bit of usage patterns where people would start GPT-IV for design and then decide to move to Essentially something cheaper, like nixtral self-hosted or nixtral self-hosted when moving to production.”
Florian Douetteau Mar 20, 2024 ▶ 20:46
Insight
Douetteau: Enterprises must treat AI and analytics pipelines as long-term IP assets
“The way I see it, enterprise need to see analytic pipelines, ML pipelines, and their apps as assets themselves. Like, IP assets they have to manage over the course of one, two, five, 10 years.”
Florian Douetteau Mar 20, 2024 ▶ 25:21
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more