Hacker
MANOHAR KUNDA
@manoharkunda
DevTrace AI
Graph HacksOct 2026
When a service fails in production, the clues are scattered: dashboards, deployment logs, past incident notes and the dependency map live in different places. On-call engineers lose time piecing together what changed and who is affected. DevTrace AI puts that context in a FalkorDB graph: services, dependencies, deployments, error signatures, incidents and confirmed resolutions. An engineer types a plain-language alert. The system finds the failing service and the error type, follows the dependency edges, and ranks candidate causes with explicit evidence tiers: a recent change on the service plus the same error seen there before gives STRONG; one of the two gives MODERATE. It shows the evidence behind each candidate, the dependency path, the blast radius (which services would feel the failure), similar past incidents with the fix that worked, and the changes it ignored and why. When evidence is thin, it says "insufficient evidence" instead of guessing. Engineers can confirm a fix. It is written back to the graph as a user-confirmed resolution and appears in the next similar investigation. Unconfirmed guesses are never saved, so the system cannot learn from its own mistakes. It is built for on-call engineers and platform teams. The core is deterministic (no LLM), uses only synthetic data, and runs locally with Docker. Backend: Spring Boot 4 on Java 25; UI: React. 75 backend tests (40 unit, 35 integration) and 3 frontend tests pass. MCP tools and live data connectors are future work.
HeatShield
Environmental HacksOct 2026
https://builder.aws.com/content/3KYnnrqEN2alcpMuo8kMtEkcUHJ/wemakedevxaws-baharth-builds-hackathon