Procurement AI Assistant
Ask questions about a public procurement dataset and inspect the answers as charts and tables.
Why I built it
I wanted to try a more conversational way to explore procurement records: ask a question such as which departments spent the most in a given year, then see the underlying results instead of only a written answer. I used a public dataset of California state purchase-order line items from fiscal years 2012–2015.
How it works
The FastAPI backend checks the question, generates a MongoDB aggregation pipeline, reviews the query result, summarizes it and suggests follow-up questions. It uses separate prompts and structured outputs for those steps. A React and TypeScript frontend sends the chat request and shows the returned data in charts and sortable tables.
Scope
This was a hobby trial, not a deployed client system. The repositories contain the backend and frontend source; running the full assistant requires the public CSV dataset, MongoDB and an OpenAI API key. I have not published a live demo or measured production accuracy.
This is a personal experiment, not a Penny Software or KPMG product. The public dataset is downloaded separately; the repositories contain no live service or included data file.