Shop-taught systems. Tools that teach. Code that survives pressure.
I ship tiny, runnable, tested public builds so regular people learning AI can enter GitHub the way a shop teaches a craft: show the job, name the hazards, leave clear notes.
| Project | What it is | Try it |
|---|---|---|
| Mister Jay | Practice the shop skills Jay taught — vehicle, electrical, plumbing, household. Illustrated Learn/Practice + Shop tip judgment. | Live |
| jay-ai-sinhala | Jay teaches friends (70+) to build with GitHub + AI — in Sinhala. Literacy before hype. | Live |
| camera-selector | DIY NVR camera selector + Frigate planning for real budgets and layouts. | Live |
One new working + tested public repo most weekdays. Pattern:
- One idea you can explain in a sentence
READMEthat teaches by running- Smoke tests that pass without secrets (mock mode)
- Clear “what you learned” notes
Roadmap (starter → mid):
| # | Repo theme | Level | State |
|---|---|---|---|
| 01 | first-commit-ai — chat CLI + mock tests | starter | shipped |
| 02 | notes-rag — retrieve, cite, answer over Markdown notes | starter | shipped |
| 03 | tiny-bpe-tokenizer — watch text become token IDs | starter | in progress |
| 04 | Tiny tool-calling agent (ReAct, no framework soup) | mid | queued |
| 05 | Prompt lab — A/B eval harness | mid | queued |
| 06 | Embedding playground (similarity you can see) | starter | queued |
| 07 | Vision caption loop | mid | queued |
| 08 | Local memory scratchpad for agents | mid | queued |
| 09 | Shop-skill explainer tied to Mister Jay drills | craft | queued |
Inspired by the spirit of small teaching repos like nanoGPT, micrograd, LLMs-from-scratch, and ai-agents-from-scratch — remixed for people who just opened GitHub.
Every teaching repo above keeps a few good first issue tickets open on purpose — scoped small, with the file to touch named in the ticket. No prior open-source experience needed.
- All open good first issues across my repos
notes-rag#1 —--jsonCLI output for answer + sourcestiny-bpe-tokenizer#1 — sample corpus atdata/sample.txtmister-jay#1 — add watch-along links to a sketch JSONattention-warrior#3 — ASCII Q/K/V attention diagram indocs/
Claim one in a comment, ask questions in the thread, ship it. Docs and diagrams count as real contributions here.
Honest reconstructed chapters (2003–2023 arc) — period-aware, meant to be run:
Map: primeodin.github.io
- Small runnable examples over grand declarations
- Honest provenance
- Teaching as proof of understanding
- Resilience before ornament
- Human judgment sets the charter; machines stop at the gates
You care about tools that teach, systems that stay clear under pressure, and builders who leave notes the next traveler can run.
Still hungry for the next winter — and the next first commit that sticks.