01 · Roasts
The 99% Solo Artist
soloPct=99 and totalPRsYear=80 — those PRs are all your own repos, aren't they? You've built a micro-empire of caveman-* tools without once leaving a comment on someone else's code.
cavekit Has 952 Stars and Zero Tests
952 people trusted cavekit enough to star it. Not one line of tests. Not one CI workflow. You wrote ARCHITECTURE.md but couldn't write `expect(ck).toBeDefined()`.
The Heatmap Tells a Story
Weeks 1–30: a ghost town of zeros. Weeks 34–52: a man possessed, hitting 4s every day. Either you discovered caffeine in August or you finally got fed up with the day job.
Naming Things Is Hard, Apparently Not
caveman, caveman-code, cavemem, cavekit, cavegemma — you have a namespace, a brand, and a theme. You also have 8 packages inside caveman-code alone. The cave deepens.
19 Years Old and 67k Stars
Your profile README says 19. Your star count says 67,653. Your follower-to-following ratio is 1577:6. What exactly are you following those 6 people for?
Built using
Zoral
Shadows one worker for a week, then takes over their job with zero extra setup. Behaves exactly like the original.
zoral.ai
02 · Category breakdown
- Impact25% weight85A
- Consistency20% weight65C
- Quality20% weight77B
- Depth15% weight72B
- Breadth10% weight65C
- Community10% weight55D
03 · Stats
365-day commit heatmap
195 active days
Language distribution
- TypeScript73%
- Swift14%
- JavaScript4%
- CSS3%
- Python2%
- MDX1%
- Other3%
04 · Numbers
Owned repos
non-fork
16
Commits
last 12 months
1,293
Followers
1,577
Joined GitHub
Apr 2022
05 · Top repos
JuliusBrussee /
caveman
Niche but well-engineered token-compression tool for AI agents. Strong production craft (CI, tests, comprehensive docs), 65k stars show real adoption. Multi-agent support + benchmark-driven development keep it shipping.
JuliusBrussee /
caveman-code
TypeScript agent framework optimizing token usage through "caveman" compression. Ships with tests, CI, full docs, 207 stars, npm package. Monorepo with 43MB codebase shows sustained architectural work across agent loops, compression middleware, multi-provider LLM support, and repomap generation.
JuliusBrussee /
cavemem
Production-ready TypeScript monorepo for persistent cross-agent memory with deterministic compression, SQLite persistence, hybrid semantic search, and IDE integrations (Claude, Cursor). 459 stars, shipped npm package, comprehensive test coverage and CI/CD.
JuliusBrussee /
pitchr
Production-ready TypeScript/Next.js SaaS with typed code, comprehensive architecture docs (ARCHITECTURE.md, design.md), CI/CD pipeline, tests, and working features (speech-to-text, pitch scoring, billing integration). 17 stars; named product with live demo at pitchr.app.
JuliusBrussee /
revu-swift
Polished local-first macOS spaced repetition app with FSRS scheduling, SwiftUI UI, SQLite persistence, and rich import/export. 301 stars, ~23 days old, active portfolio project with documented architecture and shipping product.
JuliusBrussee /
logo-designer
Pixel-grid logo editor with marching-squares contour extraction, Bézier curve fitting, and SVG/PNG export. TypeScript + React 19, full test suite, CI/CD, live GitHub Pages deployment. Competent indie web app showing clean architecture and algorithmic depth.
JuliusBrussee /
cavekit
Claude Code plugin enabling spec-driven development via compressed SPEC.md format and three slash commands (/ck:spec, /ck:build, /ck:check). Well-documented with FORMAT.md, ARCHITECTURE.md, design.md, and STATUS.md; 952 stars shows production adoption; 3.8 KB codebase with structured skills/ directory; no tests or CI,
JuliusBrussee /
JuliusBrussee
Portfolio landing page for 19-year-old developer showcasing 5+ named projects (caveman-code, cavekit, cavemem, cavegemma, pitchr). README documents multiple shipped tools with clear descriptions and links. No tests, CI, or license; minimal code artifacts provided.
JuliusBrussee /
cavegemma
Fine-tuning project for a specialized Gemma model variant using LoRA. Shipped two models to HuggingFace (merged + adapter), with evaluation metrics. Small audience (25 stars), recently created (May 2026), thin documentation, no tests/CI but well-structured training pipeline.
JuliusBrussee /
auto-karpathy
Experimental automation tool that uses Claude agents to scrape Karpathy tweets and auto-generate working Python repos. Typed, tested, and documented, but clearly a single-week sprint with minimal adoption signals.
06 · Timeline
- Apr 21, 2022Joined GitHub
- Aug 6, 2025Created JuliusBrussee
- Feb 21, 2026Created pitchr — Pitchr is an AI pitch coaching app for founder pitch practice. It supports live recording + transcription, async analysis runs, run history, deck upload/generation, and Miro fix-bo
- Mar 14, 2026Created cavekit — A Claude Code plugin that turns natural language into blueprints, blueprints into parallel build plans, and build plans into working software with automated iteration, validation,
- Mar 20, 2026Created revu-swift — Local-first macOS study app with FSRS review, decks, study guides, exams, import/export, and a polished SwiftUI desktop interface.
- Apr 4, 2026Created logo-designer — A pixel-grid logo editor that draws on a grid and outputs clean, smooth SVG/PNG — no illustrator required. Paint on a configurable grid, dial in the curve smoothing, then export pr
- Apr 4, 2026Created caveman — 🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman
- Apr 8, 2026Created caveman-code
- Apr 18, 2026Created cavemem — Cross-agent persistent memory for coding assistants. Stored compressed. Retrieved fast. Local by default.
- Apr 26, 2026Created auto-karpathy
- May 16, 2026Created cavegemma — LoRA fine-tune Gemma 4 31B to speak caveman-mode natively. Style: github.com/JuliusBrussee/caveman
- May 24, 2026Most recent push to caveman-code
07 · Compare
08 · Rubric
How this score was produced
Overall = Σ (category × weight) + gentle top-end curve
Tier thresholds
▸ How the pipeline works
- 01Scrape.Pull every non-fork repo pushed in the last 90 days, plus your contribution calendar, followers, and language byte counts — straight from GitHub's REST & GraphQL APIs.
- 02Triage.A small model reads every repo's file tree + README and picks the 20 files per repo that actually reveal how you code.
- 03Grade each repo. All repos run in parallel through a fast scoring model that reads the picked files and rates each one independently on Impact, Quality, and Depth — with evidence citations.
- 04Aggregate. A larger reasoning model combines the per-repo scores with server-computed stats (heatmap, commit cadence, language entropy, follower count) to produce the 6-dimension profile score + roasts.
- 05Correct.Deterministic server-side checks enforce anchor-scale floors (e.g. a profile with 2,000+ public commits can't score 30 Consistency) and recompute the final verdict.
~90 seconds per profile, ~$0.25 in compute. Total of ~240 files read across your top-12 repos. One rating per GitHub account per day.
▸ Data sources & caveats
- Heatmap & commit totals: GitHub GraphQL
contributionsCollection— covers the last 365 days, includes private repos when the user has opted in (default). - Language %: byte totals across the top 30 owned non-fork repos.
- Curve: a small upward nudge centered on raw score ≈ 70, capping at 100. Prevents specialists from being unfairly penalised for narrow breadth.
- Anchor corrections: when server-measured signals (e.g. privateWorkLikely, multiRepoVolume, follower count) mandate a minimum category score, the aggregation step enforces it. These are signal-conditional, not identity-based floors.