EngGauge AI

Evidence-based engineer skill assessment from GitLab & GitHub review data. Your keys, your browser, no backend.

Point EngGauge at an engineer's GitLab or GitHub handle. It reads their merged work, the comments reviewers left, and the code that drew them, then produces a weighted assessment against a target level where every claim cites a specific merge or pull request. **Your competency model, not ours.** Levels, dimensions, weights and the 1–4 behavioural anchors are all editable, exportable as JSON, and shareable across a team. The shipped ladder is a starting point. A rubric that is not yours does not fail loudly — it produces a confident, well-argued number about the wrong thing — so reports record which rubric and version produced them, and refuse to be compared across a rubric change. **There is no backend.** Your tokens live in this Chrome profile. Requests go from your browser straight to your code host and the model provider you chose. Nothing is routed through an intermediary, because there is no intermediary to route it through. With no API key set, Chrome's built-in on-device model runs a reduced analysis and nothing leaves the machine at all. **It knows what it cannot tell you.** Below a configurable evidence floor it renders the findings but withholds the level verdict — four merge requests can support "here is what reviewers keep raising" and cannot support "does not meet the bar". It flags when one repository dominates the sample. Unmeasurable dimensions score as unscored rather than as low. **Built to be argued with.** Every score links to the work it came from. Mark any merge request as unrepresentative, with a reason, and re-run without it. Export the whole report as Markdown for a growth conversation or a promo packet — the caveats travel with the conclusions. **Self-assessment mode.** The engineer can run it on themselves before their manager does, written in the second person with no verdict framing. What it needs: • A read-only GitLab or GitHub token. Self-hosted and GitHub Enterprise work. • Optionally an OpenAI, Anthropic or Gemini key for the full weighted assessment. • Optionally Confluence and Jira for documentation and delivery signal. This produces evidence, not decisions. It is a tool for preparing a conversation, not for ranking people.