AI agentsProductivity
DevTrack-AI
A weekly engineering report written from real data, where every claim cites a commit
4data sources turned into one weekly report
See the code on GitHub (opens in a new tab)
Proof you can check:Code on GitHub
AI agents
4 sources in, one honest weekly report out
- Git servercommits, issues and pull requests
- AI usagetraces, tokens, cost and speed
- Coding costsAI coding spend per project
- Numbers in codevelocity, code health, work patterns
- Weekly reportevery claim cites a commit
Lubo Bali · lubobali.com
The problem
Weekly engineering updates, for a team or a solo developer, are written from memory: what really shipped, what it cost and where the risk is gets lost.
What I built
An agent that reads commits, issues and pull requests, AI usage and costs, and coding costs, computes the numbers locally, then writes a weekly report: wins, concerns, burnout risk and what to focus on next.
Tech & why
The math is done in code first, so the AI only writes about numbers it was given; every point must cite a commit or a metric.
- Python
- NVIDIA Nemotron
- Langfuse
- Pydantic
- Forgejo
The result
A real weekly report by email from live data: velocity, code health, test discipline, AI costs and coaching, every claim traceable.
How this agent is built
Every AI agent has the same five parts. Here is what each one is in DevTrack-AI, so you can judge any agent the same way.
- ModelNVIDIA Nemotron Ultra 253B, with a Claude model as the alternative.
- HarnessCollect from 4 sources, compute the metrics locally, then let the AI write the report.
- GuardrailsPersonal data stripped before and after the AI, answers checked against a schema (retry on bad JSON), low-confidence flag when data is thin.
- KnowledgeCommits, issues, pull requests, AI traces and coding costs for the week.
- ToolsGit server API, Langfuse, a coding-cost reader and email.
Safe by design
Server addresses, keys, emails and phone numbers are removed before and after the AI sees anything, and every claim must cite a commit or a metric.
Similar projects
- AI agentsBriefMe-AIAn email chief of staff, rebuilt to do the same job with 96% fewer AI tokens96% fewer AI tokens for the same inbox run
- AI agentsSentinel-AIAn AI agent that sorts your inbox, and can never send or delete a thing4 safety layers before the AI touches an email
- AI agentsAI agents that watch their own costThree AI agents that broke in expensive ways, and the fixes that made them cheap and safe12,000× cheaper per question: $12 down to $0.001
Have a problem like this? Get your free AI blueprint in about a minute.
