AI Velocity Ledger
CLI tool that instruments a dev’s git history to estimate time-saved by AI-assisted commits and generates a weekly velocity report.
Difficulty: 1-week | Stack: Python, Click, GitPython, OpenAI API, Jinja2, SQLite
Who this is for
Engineering managers and individual devs who need to justify AI tooling spend with concrete data, mirroring what Uber finance teams would want.
Build steps
- Parse git log for commits; extract diff size (lines changed), commit message, timestamp, author
- Classify each commit with a lightweight LLM call: ‘AI-assisted’, ‘human-only’, ‘mixed’ based on message patterns and diff characteristics
- Apply a configurable velocity multiplier model (default 10x for AI-assisted) to estimate counterfactual hours
- Store results in SQLite; expose CLI commands:
ledger report --since 30d,ledger export --csv - Generate Markdown/HTML weekly report with charts (matplotlib) showing AI vs. human commit ratio, estimated hours saved, projected cost vs. tool spend
Risks
- Commit message heuristics for AI detection are noisy — teams not using conventional commit prefixes will get poor classification accuracy
- Velocity multiplier is a made-up number; report needs clear disclaimer or it becomes cargo-culted ‘proof’ in bad-faith budget meetings
- GitPython is slow on large repos with thousands of commits — needs pagination and incremental sync to SQLite
Business Angle
CLI that mines git history to quantify AI-assisted dev velocity and generates ROI reports for justifying AI tooling budgets.
Customer: Mid-level engineering manager at a 10-50 person startup who championed Cursor/Copilot/Codex adoption 6 months ago and now faces CFO asking 'what did we get for $2k/month in seats'
Pricing: one-time — $800 in month 1 via 16 x $49 one-time licenses; $300 MRR by month 3 from team-tier at $99/seat-bundle
Full business breakdown →