What was broken
Compliance documentation was assembled by hand from scattered enterprise data. The work was slow, repetitive and — because it was manual — inconsistent in exactly the place where consistency is the entire point.
Shipped
Automated Audit Reporting
An AI-driven audit report generator that extracts findings from enterprise data and drafts the compliance documentation around them, built on Azure OpenAI with semantic search underneath.
RoleAI/ML Computational Science Analyst
Duration2 years
DomainApplied NLP
StatusShipped
How it was built
I built retrieval-augmented generation on Azure AI Search and Azure OpenAI, using semantic search rather than keyword matching so that findings surfaced on meaning instead of vocabulary. NLP summarisation condenses the extracted findings into report-shaped prose. The deployment is serverless on Azure Functions, which suits a workload that runs in bursts around reporting cycles rather than continuously, with FastAPI handling the backend.
What came of it
Information retrieval accuracy improved once semantic search replaced keyword matching, and the manual effort behind each reporting cycle dropped substantially.
[00][Hours saved per cycle?]
[00][Reports generated?]
[00][Accuracy improvement?]