← Back to projects Planned · starts after Project 1 ships

RAG Compliance & Incident Intelligence Assistant

A question-answering system over UK health and safety, SIA licensing and fire safety regulations, where every answer carries citations back to the source clause. The differentiator is the evaluation harness: retrieval quality is measured in CI, not assumed.

Why this project

I have spent the last three years working in security and compliance roles alongside my studies, and I have seen how much time goes into looking up what the regulations actually require. An LLM that paraphrases regulation from memory is dangerous in that setting; an answer is only useful if it cites the exact clause it came from. That constraint, groundedness over fluency, is what this project is built around.

What it does

Architecture

IngestionDocument parsing, chunking with structure-aware splitting, metadata extraction
Indexpgvector embeddings plus BM25 keyword index
RetrievalHybrid search, then cross-encoder re-ranking
GenerationGrounded answer with inline citations; OpenAI or open-weight model
EvaluationRAGAS suite in CI: naive vs hybrid vs re-ranked, published as a comparison table

The evaluation harness

Most RAG demos show a chat box. This one ships a comparison table: the same golden question set run against naive vector search, hybrid search, and hybrid plus re-ranking, scored on faithfulness and context precision. That table is the proof that each pipeline stage earns its complexity, and it is the artefact I want to discuss in interviews.

Skills demonstrated