The akaOSS studio — open-source software for human-in-the-loop AI.
Five projects, one thesis, a reproducible research feed, and the HITL Kit component registry — served as one site at akaoss.dev.
Current benchmarks ask "can the model complete this task autonomously?" In deployment, real users want an assistant that respects their authority, preserves their agency, and makes them better over time. Assist-Not-Complete: evaluate AI on whether it assists humans without displacing them, not on whether it can finish the task alone.
The argument is made in full in the paper — An AI Measurement Problem — and tested in public in the research feed, where every finding is a reproducible experiment run with the kits below.
| Project | What it is | Status |
|---|---|---|
| HITL Kit · site | Human-in-the-loop AI, measured properly. 15 React primitives installable via the shadcn CLI (registry served from this repo), six @hitl-kit/* packages on npm — schemas, gates, LangGraph / AI-SDK / MCP adapters. |
v0.6 |
| eval-kit · site | A measurement instrument for multi-step research agents: YAML suites, per-step tool-match auto-scoring, a five-dimension human rubric, deterministic replay. Humans score, not LLMs. | v0.3.1 |
| tag-kit · site | Structured tagging primitives for annotation workflows: per-modality scoping, scope-aware agreement scoring, headless React. | stable |
| Project | What it is | Status |
|---|---|---|
| Collapse · site | A Claude Code skill-building framework — compile MDX lessons and Jupyter notebooks into SKILL.md files Claude reaches for first. |
v0.1.0 |
| Hologram · site | Live observability, guided skills, and a non-destructive MCP surface for Blender → glTF pipelines. Stdlib Python, no build step. | v0.5.0 |
This is the site repo. The projects above live in their own repos; this one holds:
src/app/ routes: / · /projects/[slug] · /research · /paper · /registry · /components
src/components/hitl/ the 15 registry primitives (source of truth for the shadcn registry)
src/components/site/ chrome: nav, footer, theming, the PixelHead mark
registry.json shadcn registry manifest → built into public/r/*.json
content/research/ findings feed posts (markdown + frontmatter)
content/paper.md An AI Measurement Problem
experiments/ reproducible experiments backing research posts (self-contained, npm-installed)
- Stack: Next.js 16 (App Router) · Tailwind v4 CSS-first · next-themes (dark default, light "warm paper",
dhotkey + header toggle) · file-based content, no CMS, no database. - The registry:
src/components/hitl/*→pnpm registry:build→public/r/*.json, served at/r/*.json. Existing consumer URLs onhitlkit.dev/r/*keep resolving via a domain alias to this site. CI fails on registry drift. - The research feed: posts in
content/research/follow a fixed shape — question, runs against real models, human-scored results, checked-in run JSON, repro link. Aggregate scores are internal signal, not leaderboard fodder.
pnpm install
pnpm dev # http://localhost:3000
pnpm verify # typecheck + registry drift check + production build
pnpm registry:build # rebuild public/r after editing registry componentsMIT © Ieuan King