AI Learning Path · Desoul Console

Teach AI fundamentals through my own shipped projects — a scrollytelling learning site fusing fundamentals + how I worked with AI, plus a live AI-news slice.

Progression Timeline Tree start roadmap →

NextRaise ep 1 from one 48-second still per beat to a treatment that actually teaches. A/B five visual arms on a single beat (panel 1.4) with narration, art and captions held constant, judged on an objective motion metric plus a comprehension judge, and ship the cheapest arm that wins. approved plan

2026-07-08shippedBuild AI Learning Path as a LOCAL-FIRST scrollytelling teaching site o…

2026-08-12shippedTurn the Studio production UI into a node canvas: every step of the co…

2026-07-08💡 ideaAI Learning Path — teach AI fundamentals through my real projects

Needs review 40

◆

0x0funky/agent-sprite-forge ◑ WATCH 40

dev-tooling · agent-skills · image-generation · sprite-generation · deterministic-postprocessing · game-assets

→ project — ai-learning-path, Comic Room panel export. more →

◆

s1dashu/animated-voiceover ◇ REFERENCE 45

other · video-generation · voiceover · tts · prompt-pipeline · content-production

◆

nolangz/pixel2motion ◇ REFERENCE 55

dev-tooling · agent-skill · logo-animation · svg · qa-gates · deterministic-validation · motion-design

○

shareAI-lab/learn-claude-code 95

claude-tooling · agent · harness · claude-code · infrastructure · learning · engineering

○

microsoft/flint-chart 75

ui-frontend · visualization · charts · ai-agents · mcp-server · multi-backend · typescript

○

QwenLM/Qwen-MM-Plugins 65

claude-tooling · multimodal · mcp-server · video-generation · document-processing · skills

○

remotion-dev/remotion ◑ WATCH 42

dev-tooling · video-generation · react · animation · rendering · batch · agent-ready

○

moonshine-ai/moonshine ◑ WATCH 20

agent-framework · voice · speech-to-text · text-to-speech · low-latency · on-device · agents

○

microsoft/VibeVoice ◑ WATCH 60

other · voice-ai · tts · asr · speech-synthesis · multilingual · realtime

○

mattpocock/dictionary-of-ai-coding ◑ WATCH 35

other · ai-terminology · reference · learning · glossary · harness-patterns · llm-fundamentals

30 more in catalog →

Apply copies one governed prompt; the session reports back and applied repos leave this queue. Undoable from history.

Plans

＋ New plan

✓

Build AI Learning Path as a LOCAL-FIRST scrollytelling teaching site on FastAPI+…

done · 8 tasks · "Built AI Learning Path: local-first FastAPI+Jinja scrollytel…"

✓

Turn the Studio production UI into a node canvas: every step of the comic pipeli…

done · 12 tasks · "Node canvas shipped: 12/12 tasks, 512 tests green. Commits f…"

◆

Raise ep 1 from one 48-second still per beat to a treatment that actually teache…

approved · 8 tasks

execution prompt →

◆

Give the cast a rigged vector body and a real motion vocabulary — point, fly, re…

approved · 9 tasks

execution prompt →

Lessons · 12 waiting · 0 kept

constitution friction proposed content-11-comic.mp4 (and the sqlite DB with per-panel timing) is gitignored and only exists in the machine that rendered it -- a fresh worktree has neither, and 'do not reach the original checkout' makes that unrecoverable without the file being handed in some other way. static/media/warroom/ in this worktree holds only content-7.mp4 and content-9*.mp4 (committed before that gitignore line existed); content-11-comic.mp4 is absent, data/*.db is absent, and the render pipeline computes per-panel duration from TTS audio length at render time, so beat 1.4's timestamp cannot even be recomputed offline.

constitution friction proposed this worktree's venv is missing numpy and PIL, which several EXISTING tests already import (test_channelart.py, half of test_comic_art.py, one in test_studio_agents.py) -- the suite starts red before any task touches it. uv run pytest on a clean worktree (before my change): test_channelart.py fails to collect (ModuleNotFoundError: numpy), plus 6 more failures across test_comic_art.py/test_studio_agents.py from the same missing PIL; confirmed pre-existing via git stash (nothing to stash) and by inspecting each failure's traceback.

approach proposed A conjunctive decision rule needs a per-axis discrimination check BEFORE it decides, or it silently returns its default and that looks like a result. The comprehension judge returned 0.000 for all five arms — one value, zero spread — so 'wins the metric AND the judge' could never promote anything; B won mechanically, not on merit. The judge was working: its reasons were specific and correct ('the claim is only present as text').

constitution friction proposed This worktree has no panel-art cache and no sqlite DB, which silently weakens any measurement that renders a real beat — it confounded the arm most dependent on the panel. All five arms rendered art-less because static/media/warroom/comic-art is gitignored and absent, and data/*.db too (init_db + WARROOM_DB into /tmp was needed just to record the gate rows). The comprehension axis then read 0.000 everywhere, biased hardest against arms A/B/D whose treatment IS the panel.

empty done proposed Task 1 was recorded done unverified: Judging harness: cut arm A from the existing mp4, build the slideshow-risk metri no verify command declared · +33361 −0

empty done proposed Task 2 was recorded done unverified: Make installed registry blocks reach the renderer — _render_work copies composit no lines changed; no verify command declared · +0 −0

empty done proposed Task 3 was recorded done unverified: Arm B — annotation only: vox-annotate + marker-highlight over the existing panel no verify command declared · +0 −9

empty done proposed Task 4 was recorded done unverified: Arm C — vector build: comparison-split ported to our tokens, rules drawing on th no lines changed; no verify command declared · +0 −0

empty done proposed Task 5 was recorded done unverified: Arm D — hybrid multi-shot: the beat becomes 4-6 shots, panel for identity, vecto no verify command declared · +0 −3

empty done proposed Task 6 was recorded done unverified: Arm E — handwritten skin of arm C using the hw-* family no verify command declared · +0 −4

empty done proposed Task 7 was recorded done unverified: Judge all five arms: layout gate, motion metric, frame grid, comprehension judge no lines changed; no verify command declared · +0 −0

empty done proposed Task 8 was recorded done unverified: Decide, record the verdict, and apply to the full episode only if a heavier arm no lines changed; no verify command declared · +0 −0

Nothing here changed a prompt until you kept it. A kept lesson retires after 90 days unless a later run re-earns it with fresh evidence.

Timeline

2026-07-08done Plan “Build AI Learning Path as a LOCAL-FIRST scrollytel…” — Built AI Learning Path: local-first FastAPI+Jinja scrollytelling site — 10 concept stops (one-paragraph explainer + click-to-expand deeper dive + a real Desoul/Financial-Analysis example each), intro narrative, live AI News slice consumed from Desoul. Cross-project task 2: added an AI News feed to Desoul Console (/ai-news + /api/ai-news, verdict=signal). Tests: ai-learning-path 4/4, research-console 125/125. Constitution: pyproject.toml created as approved scaffolding (T1); T2 edited ~/research-console per the plan (owned_paths ** permits). No blocks.

2026-08-12done Plan “Turn the Studio production UI into a node canvas: …” — Node canvas shipped: 12/12 tasks, 512 tests green. Commits ffa7911..0525936 (+ rebrand c76675a, vendored fonts 5f57401). Verified in the browser; three defects the suite could not see were caught by screenshot.

Derived from applied history + mined skills + completed plans. Ground truth stays in the project's git log; this is the index.

Outcome ● up · 7ms

http://127.0.0.1:8791

war room · revenue_cents ?war room · signups ?war room · clicks ?war room · published ?war room · backlog ?

Content room ○ not running

Send this project's material into the content pipeline. One clip is rendered first — you look at it before anything commits to a full render.

AI Learning Path is not running at http://127.0.0.1:8791

Brainstorm open space →

Think a feature or fix through with the advisor → promote to a plan.

0 open · 0 promoted · 1 done

Constitution edit rules & trust tier →

trust tier senior · custom rules · injected into every apply & execution prompt

sources 40 · 3 sub · 35 watch · 2 ref

Match rules

cat: —
tags: education · learning · learning-path · tutorials · tutorial · llm-education · ai-slides · personalized-tutoring · learning-dashboard · video-generation · video-editing · video-understanding · animation · tts · text-to-speech · speech-synthesis · voice-cloning · audio · multilingual · image-generation · visualization · multimodal

Dismissed · 33 — reviewed, no-take (restore to re-queue)

tldraw/tldraw◑ WATCH

ui-frontend

sgl-project/sglang◑ WATCH

llm-runtime

pollinations/pollinations◑ WATCH

llm-runtime

patdmp/mllm-scivis-lit-benchmark◑ WATCH

dataset

openai/whisper◑ WATCH

other

merterbak/Grok-MCP◑ WATCH