ANVIL

Compliance-grade RAG over ASME standards with auditable provenance, typed retrieval, and bounded agentic tool use.

ANVIL is a compliance-grade RAG system for pressure-vessel queries over ASME standards, built around retrieval quality, provenance, and reliable calculations.

  • Achieved 0.950 pass rate and 1.000 calculation correctness on a 100-example SPES-1 benchmark with NVIDIA NIM.
  • Combined BM25, sqlite-vec embeddings, and a typed NetworkX knowledge graph with RRF and reranking.
  • Reached 1.000 retrieval recall@10 on the benchmark.
  • Built a bounded agentic tool loop with transcript logging and guardrails, improving agent pass rate from 0.460 to 0.640 after diagnosing loop failures.

View the repository