ProofCheck is now moving from RAG into an agentic evidence-reconciliation system.
Today I built:
- 🔎 A bounded LLM investigation agent
- 🛠️ Typed tools for evidence search and validation
- 🧮 Deterministic reconciliation for quantities and calculations
- 📋 Structured outcomes: SUPPORTED / REVIEW_REQUIRED / BLOCKED
- 🛡️ Safety boundaries and controlled API retries
The core engineering principle remains:
The LLM decides what to investigate. Deterministic systems establish what the numbers actually say.
Next: Day 5 — building the interface and connecting the full workflow.
#AIEngineering #LLM #RAG #AIAgents #Python #BuildInPublic #DataScience#DecodingDataScience