Database School Building search for AI systems with Chroma CTO Hammad Bashir
29 snips
Dec 18, 2025 
Hammad Bashir, the CTO of Chroma and an expert in vector search systems, dives into the nuances of building modern search infrastructures for AI. He discusses the importance of a unified API for local and distributed systems, the evolution from local prototyping to production workflows, and the challenges of managing principled data in machine learning. Hammad also explores the implications of retrieval-augmented generation (RAG) in AI systems, as well as trends towards faster models and iterative search driven by LLMs.
AI Snips
Chapters
Transcript
Episode notes
Full-Circle Start In Vector Search
- Hammad Bashir started by building an ASIC for vector search as a high school intern and kept chasing the same problem for a decade.
- That early experience shaped his focus on retrieval and recommender systems throughout his career.
One API For Local And Cloud
- Chroma offers a single API that works both as a local embeddable DB and a distributed cloud service to match ML workflows.
- This lets developers prototype locally then promote data to production without changing code.
Explore Data Before Promoting It
- Do exploratory data analysis before tuning retrieval systems and evaluate data manually in your notebook.
- Use local iteration to curate datasets, then promote the working set to production.