DevMind - Engineering Memory Agent

Inspiration

Engineering decisions made in the past get forgotten or repeated. Teams break existing contracts, re-argue solved problems, and lose tribal knowledge when engineers leave.

What it does

DevMind is an Engineering Memory Agent that captures technical decisions and surfaces them at the right moment. When you're writing code or reviewing a PR, DevMind detects relevant past decisions and alerts you before you break them.

How we built it

Built with Python/FastAPI backend, Google Cloud Vertex AI (Gemini) for LLM reasoning, Google Agent Builder for agentic workflows, and a React/Next.js frontend. Decisions are stored in a vector database and retrieved via semantic search.

Challenges we ran into

Tuning the semantic search threshold to minimize false positives while still catching actual decision violations. Google Agent Builder's agentic chaining required careful prompt engineering.

Accomplishments that we're proud of

End-to-end pipeline from decision capture to real-time developer alerts, running entirely on Google Cloud.

What we learned

Google Agent Builder's orchestration layer dramatically reduces boilerplate for multi-step agent workflows. Vertex AI's grounding capability improves decision retrieval accuracy.

What's next for DevMind

IDE plugin (VS Code extension), Slack integration for team-wide decision broadcasting, and automated PR review bots.

Built With

• agent
• cloud
• google
• vertex