trace-field-notes / docs /submission-notes.md
JacobLinCool's picture
Add public X post link
c74d553 verified
|
Raw
History Blame Contribute Delete
2.79 kB
# Build Small submission notes
## Project
- Name: Trace Field Notes
- Track: Backyard AI
- Space: https://huggingface.co/spaces/build-small-hackathon/trace-field-notes
- Runtime: https://build-small-hackathon-trace-field-notes.hf.space/
- GitHub: https://github.com/JacobLinCool/trace-field-notes
- Demo video: https://youtu.be/1QNZlqkl8zo
- Demo MP4 asset: https://huggingface.co/spaces/build-small-hackathon/trace-field-notes/resolve/main/assets/trace-field-notes-demo.mp4
- X post: https://x.com/JacobLinCool/status/2066160425952334155
## Official pre-flight checklist
- [x] Every model is under 32B total parameters:
- `openbmb/MiniCPM5-1B`
- `nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16`
- `openai/privacy-filter`
- [x] Gradio app deployed as a Space in `build-small-hackathon`.
- [x] Demo video produced and uploaded to YouTube.
- [x] Backup demo MP4 hosted as a public Space asset.
- [x] X post published and linked from README.
- [x] ZeroGPU usage is one Space for this app.
- [x] README frontmatter includes track / quest tags and model metadata.
## Quest / challenge eligibility
- Backyard AI: eligible. The app solves a concrete workflow problem for coding
agent users.
- Best Use of Codex: eligible. Codex helped build, package, document, demo, and
verify the project; GitHub commits include Codex co-author trailers.
- Best MiniCPM Build: eligible. Quick analysis uses `openbmb/MiniCPM5-1B`.
- Nemotron Hardware Prize: eligible. Deeper analysis uses
`nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16`.
- Off Brand: eligible. The UI is a custom React field-notebook/trail-map
experience served through `gradio.Server`, not stock Gradio components.
- Best Demo: eligible. The YouTube demo, backup MP4 asset, and public X post are
prepared.
- Tiny Titan: not eligible because the project offers an optional 30B Nemotron
path.
- Best Use of Modal: not eligible; the runtime is Hugging Face ZeroGPU / CPU.
- Best Agent: not targeted. The app is agent-callable, but the app itself is not
a multi-step autonomous tool-use agent.
## Submission-form summary
Trace Field Notes is a Build Small Backyard AI app for people using coding
agents. Upload a Codex, Claude Code, or Pi Agent session log and the app turns
the agent's visible narrative messages into a qualitative field report: where it
got stuck, how it detoured, how it recovered, and whether its final claim matched
its own evidence. It uses a custom React UI served by `gradio.Server`, a
deterministic codebook analyzer, optional MiniCPM5 1B quick analysis, optional
Nemotron 3 Nano 30B-A3B deeper analysis, and privacy redaction before analysis.
Codex helped build, debug, document, deploy-package, and produce the narrated
demo video, with Codex-attributed commits in the connected GitHub repository.