**Project Overview** We need a Natural Language to SQL application that allows business users to query our data warehouse using plain English — no SQL knowledge required. The app should interpret user questions, generate accurate SQL, execute it, and return clean, meaningful results. **Tech Stack (preferred)** • LLM: Claude (Anthropic) — claude-sonnet-4-20250514 • Backend: Python + FastAPI • Database: BigQuery or Databricks (Delta Lake / Unity Catalog) • Data format: Parquet files loaded into the warehouse • Orchestration: Anthropic SDK (or LangChain if needed) **Key Requirements** • Multi-table JOIN support across a complex schema (10+ tables, millions of rows) • Schema-aware prompting — the LLM must be fed table definitions, column descriptions, and relationships to generate accurate SQL • A two-pass validation layer — generate SQL, validate it, then execute • Self-correction loop — if SQL fails, the error is sent back to the LLM for auto-correction • Safety guardrails — SELECT-only queries, row limits, injection prevention • Clean API responses that can feed a frontend or BI dashboard **Nice to Have** • Few-shot example support in the system prompt • Query explanation (show users what the SQL is doing in plain English) • Basic result visualization **Skills We're Looking For** • Strong Python / FastAPI development • Experience with LLM API integration (Anthropic or OpenAI) • Hands-on BigQuery or Databricks experience • NL-to-SQL or text-to-SQL project experience is a strong plus