🌱 Someday / Maybe
Ideas that are NOT active commitments but worth keeping. Review monthly during Weekly Review.
Overview
If an idea becomes actionable → move to TASKS-Active.
📚 Research Extensions (Post-PUMA Project)
- Add Stage 3: Backlog prioritisation (Spearman correlation ≥ 0.50)
- Add Stage 4: RAG-enhanced triage (retriever + classifier)
- Add Stage 5: Full Smart PMO multi-agent system (CrewAI or LangGraph)
- Extend to 5-class priority schema (Critical/High/Medium/Low/Trivial)
- Fine-tune Llama 3.2 on Jira SR for comparison with zero-shot
- Add fairness metrics (measure bias across project types)
- Test on non-English issue trackers (GitLab, Azure DevOps)
- Publish PUMA results as academic paper (target: MSR 2027 or EASE 2027)
- Create a live PUMA leaderboard (Hugging Face Spaces)
🛠️ Tooling Ideas
- Build Streamlit UI for interactive PUMA triage demo
- Create GitHub Action for auto-running benchmark on new model releases
- Add model-as-a-judge evaluation (use LLM to score LLM output quality)
- Integrate Arize Phoenix for full LLM observability in Stage 4+
- Try Pydantic AI for enforced structured output (replace regex parsing)
📖 Learning Topics (Post-PUMA Project)
- Deep dive: Fine-tuning LLMs (LoRA, QLoRA) — is it worth it for PM tasks?
- Learn LangGraph deeply (multi-agent coordination patterns)
- Study: AI fairness and bias measurement (Barocas, Hardt, Narayanan book)
- Explore: Program synthesis for automated spec generation (SDD extension)
💡 Open Questions Worth Investigating
- Does model size (3B vs 7B vs 13B) matter more than prompting strategy for triage?
- Can PUMA be adapted for non-Jira issue trackers (GitHub Issues, Linear)?
- What is the minimum training data needed to fine-tune a model to match CoGEE?
- Does chain-of-thought improve or degrade estimation for Fibonacci-scale outputs?
id: WAITING-Log title: ”⏳ Waiting For — PUMA” type: gtd-waiting tags: [gtd, waiting, blocked] created: 2026-03-01
⏳ Waiting For
Items where progress depends on someone/something else. Review weekly.
| Item | Waiting for | Since | Follow-up date |
|---|---|---|---|
| Tutor feedback on Ch.1 | Dr. [Tutor name] | 2026-03-08 | 2026-03-15 |
| Jira SR full dataset download confirmation | Zenodo (zenodo.org) | 2026-03-04 | Verify SHA256 |
| TAWOS dataset download | GitHub SOLAR-group | 2026-03-04 | ✅ Done |
| Ollama v0.5+ release (seed API fix) | Ollama project | TBD | Check monthly |
id: PROJECTS-Master title: ”📋 Projects Master List — GTD” type: gtd-projects tags: [gtd, projects, master] created: 2026-03-01
📋 Projects Master List
All active multi-step projects. Each must have a clear next action.
| Project | Next Action | Deadline | Note |
|---|---|---|---|
| PUMA Project — Ch.2 Literature | Run SLR search on arXiv | 2026-04-08 | PR-PUMA-Ch2-Ch3-Ch4-Ch5 |
| PUMA Project — Triage Module | Write SP-Triage-Agent-v1 spec | 2026-04-08 | SP-Architecture |
| PUMA Vault — GitHub | Set up git remote | 2026-03-06 | README |
| Zotero Library | Import 6 verified papers | 2026-03-08 | 20 - Literature |
| Ollama Environment | Pull Mistral 7B + verify | 2026-03-05 | LN-Tools-Ollama-ClaudeCode-OpenCode-BrowserOS |