Audit, clean up, and deploy a Read-Along AI Gradio app to a Hugging Face Space
Session from read-along-ai-agent-traces on Hugging Face by kingkw1 · MIT · condensed by Coders Talk
1h 28msession
0interventions
GOAL
Audit the read-along-ai hackathon project against its evolving vision and eligibility rules, clean up docs/tests, then deploy the app as a Hugging Face Space and fix the runtime issues that came up.
- +0Prompt · +0Audit our docs, readme, and app against the hackathon vision and eligibility, including badges/awards; we dropped the multi-reading-level idea.
- +0mAgent did · +0mRead README, specs, app.py, inference modules and tests; ran the light pytest suite (4 failures from stale output contracts); produced an audit summary of drift, eligibility, and repo size (16GB with GGUF models).
- +11mPrompt · +11mUpdate spec docs and README for the sentence-only MVP, keep placeholders and Off the Grid/Llama Champion claims, and clean up stale tests into sensible folders.
- +11mAgent did · +11mMoved tests into tests/, converted local smoke test to a manual script, updated docs/README/requirements to reflect the sentence MVP, split requirements-dev.txt, and fixed path/import breakage until pytest was green (32 passed, 2 skipped).
- +37mPrompt · +37mWalk me through creating the Hugging Face Space and review hackathon submission guidelines before doing more UI tweaks.
- +41mFail · +41mgit push to the Space remote was rejected, then HF CLI upload failed with 403 because the saved token only had read permission.
- +42mPrompt · +42mTake this over yourself until you hit something only I can fix, like credentials.
- +42mAgent did · +42mDiagnosed the credential wall, staged a vetted small upload folder (excluding .hackathon, models, caches) and, once a write token was supplied, pushed via hf upload successfully.
- +51mFail · +51mFirst Space build failed: Python 3.13 default plus pinned transformers/huggingface-hub versions conflicted with Gradio 6's requirements.
- +52mAgent did · +52mPinned python_version 3.12, made requirements lean and Modal-first, fixed local-vs-Modal prewarm routing, guarded missing Modal secrets, fixed a growing-whitespace CSS layout bug, iteratively re-uploaded and polled the Space until it ran cleanly, and built a reusable deploy_space.sh script as the standard upload shortcut.
- Break · 5h 09m away, not counted
- +1h 28mOutcome · +1h 28mSpace build clean and RUNNING with a live Modal TTS smoke test passing; local suite at 36 passed, 2 skipped; repo restructured with a repeatable deploy_space.sh upload shortcut.
VERDICT
Next time I'd ask upfront for a deploy script instead of letting the agent hand-run git/HF CLI commands each time — that's what actually stabilized the workflow. The correction that mattered most was catching the read-only HF token and the Python 3.13/Gradio 6 dependency conflict; once those were fixed the Space just worked. I'd also front-load 'don't upload local model files or .hackathon notes' so I don't have to catch that mid-session.