prisma-chatbot / ROADMAP.md
RolandM's picture
Finalize README, ROADMAP, and env template for launch
9fdc8c3
# Roadmap
Deployment plan and milestones for prisma-chatbot. Living document — items
move from *Planned* to *In progress* to *Done* as the project evolves.
## Milestone 1 — Scaffolding
> **Done:** Repo skeleton, docs, dependency manifest, env template,
> gitignore. No chatbot logic yet.
- [x] Repo created, license added
- [x] CLAUDE.md, README, ARCHITECTURE, ROADMAP drafts
- [x] `src/`, `tests/`, `assets/` directories with placeholder modules
- [x] `requirements.txt`, `.gitignore`, `.env.example`
## Milestone 2 — Minimal end-to-end loop
> **Done:** Get a single message in / response + evaluation out working
> locally, even with a rough prompt.
- [x] Implement `src/config.py` (model id, attributes, turn cap)
- [x] Implement `src/prompt.py` (v1 dual-role prompt)
- [x] Implement `src/inference.py` (HF Inference client wrapper)
- [x] Implement `src/evaluation.py` (JSON parsing + validation)
- [x] Implement `app.py` (minimal Gradio UI)
- [x] First pytest tests for parsing/validation
## Milestone 3 — Prompt and UX iteration
> **Done:** Refine Prisma's voice, evaluation display, and the "check what
> I think" affordance.
- [x] Refine system prompt for voice consistency and structured-output
reliability
- [x] Decide evaluation display: numeric, verbal, or both
- [x] Decide update cadence: live each turn vs. on-request
## Milestone 4 — Public deployment
> **Done:** Shipped to a Hugging Face Space and linked from the README.
- [x] Hugging Face Space configuration
- [x] Per-session turn cap
- [x] Public URL added to README and papers
## Milestone 5 — Stretch ideas
> **Planned:** Explicitly non-blocking; consider only after the demo is live
> and stable.
- [ ] About panel copy (`assets/about.md`)
- [ ] Rate limiting (per-IP / per-session beyond the turn cap)
- [ ] Light usage analytics (anonymous, aggregate)
- [ ] "Compare models" mode
- [ ] Downloadable conversation + evaluation transcript
- [ ] Linguistic feature highlighting (which words/choices shifted scores)
## Future work (v2+)
### Attribute customization
- User selects up to 6 attributes from a curated extended list (~15–20
dimensions covering social, cognitive, affective, and pragmatic
perception)
- Extended list includes: pushy, knowledgeable, well-prepared, pedantic,
helpful, arrogant, warm, evasive, confident, anxious, etc. (final set TBD)
- v1's default set remains as the "quick start" option