Energy Copilot
Python - FastAPI - PostgreSQL - pgvector - Next.js
In progress: an AI copilot that answers plain-English questions over public Alberta well production data.
Building an AI copilot that answers plain-English questions over public Alberta well production data (Petrinex, AER). This project is in progress.
Wrote the product spec: target users, priority use cases, success metrics and a phased roadmap.
Using a tool-calling agent that writes guarded read-only SQL, forecasts decline curves, and cites AER directives, with pgvector retrieval for RAG.
Designed responsible-AI guardrails: answers grounded in query results, cited regulations, and refusals on confidential data.
Developing eval-first: versioned test sets for SQL accuracy, retrieval and extraction, used to compare RAG chunking and hybrid retrieval strategies, with CI blocking regressions.