Discover
Work directly with the client to understand the bid-automation opportunity.
03 · Client product · RAG · Construction
Turning a broad AI opportunity into a production tool.
For a real construction client, our four-person capstone team converted an AI opportunity into scoped requirements and a GPT-4 and ChromaDB product that the client adopted for production use.
4person delivery team
GPT-4 + RAGproduct architecture
Productionclient adoption outcome
The product task
Translate “use AI for bids” into a product a real client could adopt.
The project began with a real construction client and a broad automation opportunity. I ran client discovery, helped scope requirements, and worked with a four-person team to deliver a RAG-based bid-automation tool.
The implementation combined GPT-4 with ChromaDB. The most important published outcome is not a model benchmark: it is that Acres adopted the product for production use.
From ambiguity to delivery
A compact view of the product work that is documented.
Scope
Turn the opportunity into requirements a four-person team could deliver.
Ship
Deliver the GPT-4 and ChromaDB tool that the client adopted in production.
Why this belongs in a PM portfolio
The transferable behaviour is product scoping—not an inflated title.
Real stakeholder discovery
The requirements came from a live client context rather than a speculative demo.
A defined technical approach
RAG, GPT-4, and ChromaDB gave the team a concrete delivery boundary.
Adoption over presentation
The client’s production adoption is the strongest available signal of usefulness.
Limits of the public evidence
This is intentionally a concise case study.
The current résumé does not provide usage volume, time saved, decision alternatives, or the failed assumption identified for a fuller retrospective. I am not filling those gaps with invented detail.
Details to add
Client-approved artifacts, the central product trade-off, what did not work, and post-adoption measurement.
What I would measure next
“Production adoption is a strong beginning, but it is not the complete product outcome.”
A stronger follow-up would track task time, edit rate, output acceptance, repeated use, and the bid stages where the product creates—or fails to create—meaningful value.