Building With AI

Boosting the productivity of data scientists with AI | Bryan Bischof (AI Lead at Hex)

Feb 12, 2024

Bryan Bischof, AI Lead at Hex and author of "Building Recommendation Systems," shares his expertise on enhancing data science workflows through AI. He discusses the limitations of synthetic data and details his innovative three-stage recommender system. Bryan sheds light on ineffective retrieval techniques, Hex's groundbreaking metrics for AI performance, and the unique user-centric approach the company takes. Plus, he reveals intriguing interview questions as Hex looks to expand its talent pool!

01:02:31

Creator website

Episode guests

Bryan Bischof

AI Summary

AI Chapters

Episode notes

Podcast summary created with Snipd AI

Quick takeaways

Integrating LLMs into data science workflows can significantly enhance data analysis processes, shifting away from traditional methods to innovative solutions.

Prioritizing user needs using the Jobs To Be Done framework ensures that AI features effectively contribute to improving the overall user experience.

Effective retrieval strategies are crucial for enhancing AI performance, involving not just relevant data identification but also result re-ranking to address complex queries.

Deep dives

Inception of HexMagic

HexMagic serves as an AI co-pilot designed specifically for data scientists. Its inception was driven by the need to enhance existing data science workflows by integrating advanced AI capabilities. The transition toward using AI was influenced by Brian’s previous experiences in the data field, including his challenges with traditional data visualization tools. Ultimately, he recognized that LLMs could significantly improve the data analysis processes at Hex by providing augmented capabilities, marking a shift from conventional methods to more innovative solutions.

Intro

2min

Enhancing Data Science Workflows with AI Innovations

4min

Enhancing Data Science with AI and Hex

30min

Unlocking Modern Recommendation Systems

14min

Exploring the Benefits of Hex for Prototyping

2min

Integrating AI: Metrics and Management

9min

Upcoming Features and Team Expansion at XMagic

2min

In this episode of Building With AI, host Haroon Choudery is joined by Bryan Bischof, the AI lead at Hex. Bryan joined the Hex Magic team in 2023, leveraging his insights to improve data science workflows through LLM integrations.

During the conversation, they discuss:

The Hex way of approaching jobs to be done,
Why Bryan is over synthetic data,
Bryan's three-stage recommender system,
Retrieval techniques that don't work,
Hex's game-changing metrics for AI performance,
And much more!

Bonus: Bryan reveals the questions he asks candidates during interviews...and guess what? Hex is currently on the lookout for fresh talent! So keep your 👀 peeled!

—

[00:00] Up next:

[00:08] Episode introduction

[00:55] Autoblocks: AI Optimization Platform

[01:48] Bryan’s introduction

[06:47] The Hex way of approaching jobs to be done

[10:34] To blank slate or not to blank slate

[11:32] How Hex prioritizes different pain points

[15:37] Challenges of building Hex: chain reactions, entropy, and latency issues

[19:04] How to build accurate and reliable agents

[22:27] The solution to evals

[26:33] How to build powerful evals

[31:46] Why Bryan is over synthetic data

[35:48] Bryan's three stage recommender system

[44:35] The role of data managers

[46:27] Retrieval techniques that don't work

[47:57] The team behind Hex Magic

[50:04] Hex vs. Jupyter

[51:41] Hex's game-changing metrics for AI performance

[54:39] The nuances of developing AI products

[01:00:40] Hex's future projects

—

Brought to you by Autoblocks (https://www.autoblocks.ai/) — how product teams deliver better AI experiences.

—

Building with AI is a weekly podcast featuring conversations with top AI product builders and leaders. Through thoughtful dialogue, host Haroon Choudery (CEO of Autoblocks) uncovers tactical insights that help listeners build & scale world-class AI products.

—

Subscribe to Building With AI: