DANIEL KIRCHNER
All projects

AI Audit Assistant for Certification & Compliance

Architecture and agent design

PythonLangGraphFastAPIAzure OpenAIpgvectorNext.js

Problem

An assistant for certification audits. Auditors write raw notes. Those notes have to become structured findings that match the standards. Free text without a source is useless in an audit.

Constraints

Industry: certification and compliance. Every claim needs a clause in the standard. No client name, no auditor contacts.

Architecture

Several LangGraph agents, RAG over the standards, a source on every claim. Azure OpenAI, pgvector, FastAPI, Next.js.

Evaluation

Without source grounding the output is useless in an audit. A finding without a clause counts as an error, not a draft.

Architecture
Rohnotizen
→
LangGraph-Agenten
→
RAG über Normen
→
Finding + Stelle

extrahieren · belegen · zuordnen

Beta. Findings tied to the standard, not free text nobody can check.