StarZero All shows Try the podcast clipper →

Apr 5, 2021 · 30m · mad

Fireside Chat: Bindu Reddy (Founder & CEO, Abacus.AI) with Matt Turck (Partner, FirstMark)

Bindu Reddy · 22m spoken Matt Turck · 5m 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

In this fireside chat from Data Driven NYC, Abacus.AI Founder & CEO Bindu Reddy joins FirstMark Partner Matt Turck to discuss deep learning fundamentals, the architecture of end-to-end autonomous AI platforms, and practical strategies for machine learning adoption in startups and enterprises.

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 19% of the talking time here. How this is scored →

Matt as informed peer 3.0 Guest teaching 3.4 Guest disagreement 0.6 Matt pushing back 0.4
05100:0010:0020:0030:000:10–5:43 · Matt as informed peer 2/10 Defining Neural Networks and Learning Types Matt opens the episode by setting up introductory questions about neural networks and learning types. Bindu provides extensive educational background on nodes, weights, biases, and AlexNet, taking on the role of teacher.5:43–8:54 · Matt as informed peer 2/10 Applying Deep Learning to Tabular Data Matt prompts Bindu to explain how deep learning applies to tabular data. Bindu details predictive modeling examples like Zillow estimates and explains why neural nets excel with high-dimensional data over tree algorithms.8:54–11:24 · Matt as informed peer 3/10 Abacus.AI and End-to-End Autonomous AI Matt asks about Abacus.AI's mission and interjects to note he was just about to ask about AutoML right before Bindu brings it up. Bindu clarifies how Abacus handles full data processing end-to-end beyond traditional AutoML.11:24–14:04 · Matt as informed peer 4/10 Technical Architecture and Product Roadmap Matt demonstrates technical understanding by accurately guessing that Abacus builds glue around open-source ML frameworks. Bindu enthusiastically validates his premise and outlines their engineering stack.14:04–18:55 · Matt as informed peer 3/10 Target Customers and Feature Stores Explained Matt identifies 'feature stores' as a current industry hot topic and asks for a definition. Bindu gives a comprehensive explanation using a Netflix user session example to show how feature stores eliminate redundant data wrangling.18:55–24:13 · Matt as informed peer 4/10 Startup ML Adoption and Deep Learning Mastery Matt brings research to the interview by quoting several of Bindu's provocative tweets on ML adoption and experimentation. Bindu elaborates on self-directed learning and curiosity in ML engineering.24:13–27:55 · Matt as informed peer 3/10 Audience Q&A: Data Labeling, DNN Models, and Competition Matt fields audience questions, inviting Bindu to highlight advantages over competitors like H2O and DataRobot. Bindu reframes these companies as non-competitors focused solely on AutoML rather than full data wrangling.0:10–5:43 · Guest teaching 5/10 Defining Neural Networks and Learning Types Matt opens the episode by setting up introductory questions about neural networks and learning types. Bindu provides extensive educational background on nodes, weights, biases, and AlexNet, taking on the role of teacher.5:43–8:54 · Guest teaching 4/10 Applying Deep Learning to Tabular Data Matt prompts Bindu to explain how deep learning applies to tabular data. Bindu details predictive modeling examples like Zillow estimates and explains why neural nets excel with high-dimensional data over tree algorithms.8:54–11:24 · Guest teaching 3/10 Abacus.AI and End-to-End Autonomous AI Matt asks about Abacus.AI's mission and interjects to note he was just about to ask about AutoML right before Bindu brings it up. Bindu clarifies how Abacus handles full data processing end-to-end beyond traditional AutoML.11:24–14:04 · Guest teaching 2/10 Technical Architecture and Product Roadmap Matt demonstrates technical understanding by accurately guessing that Abacus builds glue around open-source ML frameworks. Bindu enthusiastically validates his premise and outlines their engineering stack.14:04–18:55 · Guest teaching 5/10 Target Customers and Feature Stores Explained Matt identifies 'feature stores' as a current industry hot topic and asks for a definition. Bindu gives a comprehensive explanation using a Netflix user session example to show how feature stores eliminate redundant data wrangling.18:55–24:13 · Guest teaching 2/10 Startup ML Adoption and Deep Learning Mastery Matt brings research to the interview by quoting several of Bindu's provocative tweets on ML adoption and experimentation. Bindu elaborates on self-directed learning and curiosity in ML engineering.24:13–27:55 · Guest teaching 3/10 Audience Q&A: Data Labeling, DNN Models, and Competition Matt fields audience questions, inviting Bindu to highlight advantages over competitors like H2O and DataRobot. Bindu reframes these companies as non-competitors focused solely on AutoML rather than full data wrangling.0:10–5:43 · Guest disagreement 0/10 Defining Neural Networks and Learning Types Matt opens the episode by setting up introductory questions about neural networks and learning types. Bindu provides extensive educational background on nodes, weights, biases, and AlexNet, taking on the role of teacher.5:43–8:54 · Guest disagreement 0/10 Applying Deep Learning to Tabular Data Matt prompts Bindu to explain how deep learning applies to tabular data. Bindu details predictive modeling examples like Zillow estimates and explains why neural nets excel with high-dimensional data over tree algorithms.8:54–11:24 · Guest disagreement 1/10 Abacus.AI and End-to-End Autonomous AI Matt asks about Abacus.AI's mission and interjects to note he was just about to ask about AutoML right before Bindu brings it up. Bindu clarifies how Abacus handles full data processing end-to-end beyond traditional AutoML.11:24–14:04 · Guest disagreement 0/10 Technical Architecture and Product Roadmap Matt demonstrates technical understanding by accurately guessing that Abacus builds glue around open-source ML frameworks. Bindu enthusiastically validates his premise and outlines their engineering stack.14:04–18:55 · Guest disagreement 0/10 Target Customers and Feature Stores Explained Matt identifies 'feature stores' as a current industry hot topic and asks for a definition. Bindu gives a comprehensive explanation using a Netflix user session example to show how feature stores eliminate redundant data wrangling.18:55–24:13 · Guest disagreement 1/10 Startup ML Adoption and Deep Learning Mastery Matt brings research to the interview by quoting several of Bindu's provocative tweets on ML adoption and experimentation. Bindu elaborates on self-directed learning and curiosity in ML engineering.24:13–27:55 · Guest disagreement 2/10 Audience Q&A: Data Labeling, DNN Models, and Competition Matt fields audience questions, inviting Bindu to highlight advantages over competitors like H2O and DataRobot. Bindu reframes these companies as non-competitors focused solely on AutoML rather than full data wrangling.0:10–5:43 · Matt pushing back 0/10 Defining Neural Networks and Learning Types Matt opens the episode by setting up introductory questions about neural networks and learning types. Bindu provides extensive educational background on nodes, weights, biases, and AlexNet, taking on the role of teacher.5:43–8:54 · Matt pushing back 0/10 Applying Deep Learning to Tabular Data Matt prompts Bindu to explain how deep learning applies to tabular data. Bindu details predictive modeling examples like Zillow estimates and explains why neural nets excel with high-dimensional data over tree algorithms.8:54–11:24 · Matt pushing back 1/10 Abacus.AI and End-to-End Autonomous AI Matt asks about Abacus.AI's mission and interjects to note he was just about to ask about AutoML right before Bindu brings it up. Bindu clarifies how Abacus handles full data processing end-to-end beyond traditional AutoML.11:24–14:04 · Matt pushing back 0/10 Technical Architecture and Product Roadmap Matt demonstrates technical understanding by accurately guessing that Abacus builds glue around open-source ML frameworks. Bindu enthusiastically validates his premise and outlines their engineering stack.14:04–18:55 · Matt pushing back 0/10 Target Customers and Feature Stores Explained Matt identifies 'feature stores' as a current industry hot topic and asks for a definition. Bindu gives a comprehensive explanation using a Netflix user session example to show how feature stores eliminate redundant data wrangling.18:55–24:13 · Matt pushing back 1/10 Startup ML Adoption and Deep Learning Mastery Matt brings research to the interview by quoting several of Bindu's provocative tweets on ML adoption and experimentation. Bindu elaborates on self-directed learning and curiosity in ML engineering.24:13–27:55 · Matt pushing back 1/10 Audience Q&A: Data Labeling, DNN Models, and Competition Matt fields audience questions, inviting Bindu to highlight advantages over competitors like H2O and DataRobot. Bindu reframes these companies as non-competitors focused solely on AutoML rather than full data wrangling.

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

0:00 · Matt 23.7% · guest 76.3%0:00 · Matt 23.7% · guest 76.3%3:00 · Matt 10.2% · guest 89.8%3:00 · Matt 10.2% · guest 89.8%6:00 · Matt 3.7% · guest 96.3%6:00 · Matt 3.7% · guest 96.3%9:00 · Matt 12.1% · guest 87.9%9:00 · Matt 12.1% · guest 87.9%12:00 · Matt 14.7% · guest 85.3%12:00 · Matt 14.7% · guest 85.3%15:00 · Matt 3.1% · guest 96.9%15:00 · Matt 3.1% · guest 96.9%18:00 · Matt 32.1% · guest 67.9%18:00 · Matt 32.1% · guest 67.9%21:00 · Matt 26.9% · guest 73.1%21:00 · Matt 26.9% · guest 73.1%24:00 · Matt 21.4% · guest 78.6%24:00 · Matt 21.4% · guest 78.6%27:00 · Matt 30.6% · guest 69.4%27:00 · Matt 30.6% · guest 69.4%30:00 · Matt 94.3% · guest 5.7%30:00 · Matt 94.3% · guest 5.7%
Sharpest disagreement ▶ 26:26 Rejection of competitor framing

When Matt asks her to compare Abacus to H2O and DataRobot, Bindu explicitly rejects the premise, stating she views them as non-competitors because they only focus on AutoML.

 Hardest push from Matt ▶ 10:40 Matt asserts question timing on AutoML

As Bindu transitions to mention AutoML, Matt interrupts to state 'I was about to ask,' asserting host control and demonstrating he was tracking the topic ahead of time.

 Biggest teaching moment ▶ 16:27 Detailed explanation of ML feature stores

After Matt asks for a definition of feature stores, Bindu breaks down petabyte-scale data transformation challenges and explains how feature stores function using a Netflix user calculation example.

 Matt holds his own ▶ 11:22 Host accurately diagnoses system architecture

Matt demonstrates technical domain insight by correctly predicting that Abacus relies on open-source frameworks tied together with custom glue code, receiving instant validation from the guest.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Defining Neural Networks and Learning Types 2500 Matt opens the episode by setting up introductory questions about neural networks and learning types. Bindu provides extensive educational background on nodes, weights, biases, and AlexNet, taking on the role of teacher.
Applying Deep Learning to Tabular Data 2400 Matt prompts Bindu to explain how deep learning applies to tabular data. Bindu details predictive modeling examples like Zillow estimates and explains why neural nets excel with high-dimensional data over tree algorithms.
Abacus.AI and End-to-End Autonomous AI 3311 Matt asks about Abacus.AI's mission and interjects to note he was just about to ask about AutoML right before Bindu brings it up. Bindu clarifies how Abacus handles full data processing end-to-end beyond traditional AutoML.
Technical Architecture and Product Roadmap 4200 Matt demonstrates technical understanding by accurately guessing that Abacus builds glue around open-source ML frameworks. Bindu enthusiastically validates his premise and outlines their engineering stack.
Target Customers and Feature Stores Explained 3500 Matt identifies 'feature stores' as a current industry hot topic and asks for a definition. Bindu gives a comprehensive explanation using a Netflix user session example to show how feature stores eliminate redundant data wrangling.
Startup ML Adoption and Deep Learning Mastery 4211 Matt brings research to the interview by quoting several of Bindu's provocative tweets on ML adoption and experimentation. Bindu elaborates on self-directed learning and curiosity in ML engineering.
Audience Q&A: Data Labeling, DNN Models, and Competition 3321 Matt fields audience questions, inviting Bindu to highlight advantages over competitors like H2O and DataRobot. Bindu reframes these companies as non-competitors focused solely on AutoML rather than full data wrangling.

Statements from this episode (8)

Assertion Not checkable as stated
95% of machine learning models at Google and Facebook are neural nets
“So if you look at a company like Google or Facebook, almost all of their models, like, 95% of their models are neural net based.”
Bindu Reddy Apr 5, 2021 ▶ 7:12
Insight
Data processing and wrangling is the most problematic part of machine learning
“AutoML is one piece of this, which is finding the right model based on your data. We actually automate the whole end-to-end, and most of the chunk of the work we do is in the data processing wrangling management, which, you know, if you guys are a data scienti…”
Bindu Reddy Apr 5, 2021 ▶ 10:58
Assertion Supported
An Abacus.AI co-founder is the original creator of Google BigQuery
“One of our co-founders is the founder of BigQuery.”
Bindu Reddy Apr 5, 2021 ▶ 12:18
Assertion Partly supported
Pinterest uses AWS exclusively for its cloud infrastructure
“Pinterest of course only uses AWS, as they've mentioned.”
Bindu Reddy Apr 5, 2021 ▶ 15:52
Insight
10,000 data points is the threshold to start building machine learning models
“If you have 10,000 data points, that's my rule of thumb. If you have 10,000 data points, you're ready to build a model.”
Bindu Reddy Apr 5, 2021 ▶ 21:00
Prediction Not checkable as stated
Machine learning models for images and language will soon require less data
“So, ah, you know, I think the state of the art is moving very rapidly, ah, and so that you can, like, teach models with less and less data, and this will be the case for both images and language very soon.”
Bindu Reddy Apr 5, 2021 ▶ 25:15
Opinion
Bindu Reddy considers H2O.ai and DataRobot to be non-competitors to Abacus.AI
“So if somebody actually knows our space, but HTO and data robot are really, we, I mean, maybe they consider themselves our competitors. We like to think of them as non-competitors. The reason for that is their real focus, as you probably may know, is on the au…”
Bindu Reddy Apr 5, 2021 ▶ 26:31
Prediction Not checkable as stated
Real-time transformation tools for large data volumes are the next big trend
“Anything and everything which helps you do data transformations on large amounts of data in real time, I think is going to be the big thing, and sooner or later we're going to figure something out there.”
Bindu Reddy Apr 5, 2021 ▶ 29:46
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