SDSU Software Procurement Assistant

An AI rebuild of how SDSU reviews software requests. Chatbot intake, automatic compliance flagging, and AI-generated risk reports, running on a serverless AWS pipeline.

React
Python
AWS Lambda
Amazon Bedrock
DynamoDB
AWS Amplify

View on GitHub

Contributors: 6
Built in: under 3 days

Problem

Every software request at SDSU needed a mandatory interview with IT staff just to translate the technical questions into plain English. Whatever the reviewer found lived in tickets and spreadsheets with no way to search past decisions, and requesters had no idea where their request even stood.

What we built

Requesters fill out an intake form, then talk to an AI chatbot (built on Amazon Bedrock) that basically replaces the interview a human reviewer used to run. The system flags requests for ATI, Security, or Integration review automatically, using the same thresholds a reviewer would apply by hand, then puts together a risk report by actually reading the vendor's privacy policy, Terms of Service, VPAT, and HECVAT, pulling them off the web itself if nobody attached them already. The catalog it matches against is seeded from 798 real SDSU software titles, deduped from the university's own ServiceNow records.

Architecture

Everything's serverless on AWS. A React (Vite) frontend on Amplify hits a REST API through API Gateway, which routes to Python Lambda functions. Lambda reads and writes DynamoDB, calls Bedrock for the chatbot and the risk reports, and drops supporting documents in S3. SES handles every notification along the way, confirmations, missing-document nags, the final verdict.

Performance

Report generation had to run async because a real VPAT actually took 126 seconds to process through Bedrock, and API Gateway cuts you off at 29 seconds for a normal request, that's over 4x longer than the platform allows in one shot. So the endpoint kicks off a background worker and returns right away instead of blocking. Security reports came back faster, around 24 seconds. By the time we wrapped up the build, 116 tests were passing (65 backend, 51 frontend).

Context

Built with a team at Cal Poly's DxHub AI Summer Camp, powered by AWS, as a project for SDSU.