Skip to content
All projects

2026 · 1st · ZeroDay MLH

DataMind

Natural-language data agent where every answer carries its evidence: the executed SQL, bound parameters, result rows and Snowflake query ID.

Source

What I built

  • Open-source natural-language data agent. Won 1st place in the Open Source Track at the ZeroDay MLH national hackathon (2026).
  • Google Gemma, running locally via Ollama, turns a question into a constrained semantic plan; deterministic code validates it and compiles it into parameterized, read-only, allowlisted SQL.
  • Snowflake computes the answer, and every response ships with its evidence: executed SQL, bound parameters, result rows and the Snowflake query ID.
  • Semantic YAML data model and a CI quality gate keep the agent auditable and reproducible.

Stack

PythonGemmaOllamaSnowflakeSQL