The Joe Reis Show Simba Khadder - Feature Stores, Reinforcement Learning, and More
23 snips
Jan 2, 2025 Simba Khadder, Founder and CEO of FeatureForm, shares his expertise in building feature stores and large-scale recommender systems. He discusses the evolution and significance of feature stores in machine learning, the challenges of applying large language models, and innovative data management practices with Iceberg. Simba also reflects on the transformative impact of AI tools on coding and the tension between modern technology and tradition. Plus, he offers insights on surfing as a mindfulness practice, sharing personal stories from his adventures.
AI Snips
Chapters
Transcript
Episode notes
ML in Production
- Production machine learning resembles data engineering more than traditional data science.
- The shift requires a change in approach, moving from experimental notebooks to robust production pipelines.
LLMs vs. Traditional ML
- LLMs are in an experimental phase, with companies still exploring optimal applications beyond chatbots.
- Traditional machine learning models remain essential for specific use cases like fraud detection.
LLM Limitations
- Khadder gave an example of a bank chatbot's limitations in accessing structured data like credit scores.
- LLMs often rely on text data in vector databases, neglecting information stored in traditional databases.