Multi-Agent PR Reviewer
A working multi-agent application, not a demo. The browser fetches the PR + file diffs from the GitHub public API, then fans out four parallel streaming LLM calls — each with a tightly scoped system prompt focused on a single review dimension (correctness · security · style · tests). Cards stream in independently as the model produces tokens. Once all four finish, a lead-reviewer prompt receives the diff plus all four reviews and emits a structured verdict (LGTM / LGTM_WITH_NITS / NEEDS_CHANGES / BLOCK) plus top severity-graded issues. Reuses the existing OpenRouter streaming route, the per-IP rate limiter, and the same UI patterns as the rest of AI Labs. Lives at /pr-review on the portfolio.
Built in public to learn something or prove a pattern. Some are published packages, some are demos that live on this site.
- ▸Most 'AI code review' tools are single-call wrappers — one prompt does everything, blurring critique across dimensions
- ▸GitHub diffs can be enormous; sending the whole thing to a model wastes context and money
- ▸Multi-step agent flows tend to feel slow and opaque if the UI doesn't show the work happening
- ▸Four parallel specialised reviewer agents, each with a single-dimension system prompt — cleaner critique, smaller per-call context, half the latency of a sequential pipeline
- ▸Server-side PR fetch ranks files by total changes and keeps the top 10 with per-file diff truncation; total context stays well under 50 KB
- ▸Live streaming cards per reviewer + a final structured-verdict card so the visitor sees parallel orchestration happening in real time
Working tool that reviews real public PRs in seconds
Demonstrates real multi-agent orchestration — not a sandbox simulation
Reuses /api/demo/generate, the shared rate limiter, and the AI Labs UI vocabulary
JobJam.io — AI Job Search & Application Platform
AI-powered job search platform: discover roles, evaluate fit, tailor applications, and close skill gaps. One-time pricing, no subscriptions.
Fathohm — Comprehension-Debt System of Record
See how much of your codebase no human understands. A treemap of your repo colored by human comprehension score, with a tracked, assignable metric for 'who understands what' in an AI-native codebase.
Grenz — Firewall for AI Agents
A local firewall for AI agents: scoped, revocable, observable permissions instead of your full GitHub token. Deny-by-default, MIT licensed, live at grenz.dev.