Summary
AI researcher, creator of DSPy — a framework for programming (rather than prompting) LLM pipelines. Co-author of Meta-Harness, which automates the search for optimal harness configurations.
Key Points
- Created DSPy, which treats LLM pipelines as optimizable programs rather than fixed prompt templates
- DSPy’s philosophy aligns with Meta-Harness: both view LLM system configuration as a search/optimization problem rather than a manual engineering task
- Meta-Harness extends this further by searching over arbitrary code (not just prompt modules)
Evidence Timeline
- 2026-04-07: Co-author of Meta-Harness paper (arXiv:2603.28052)