AI Internship — Todo List
capstone ai-internship neurology todo
Master todo list for the capstone. Updated weekly. Check off items as completed.
Week 1-2 — Research Phase ✅ DONE
Completed
- Neurology AI landscape map (16 competitors across 5 layers)
- 34 academic papers tiered (5 priority reads)
- 12+ Reddit patient pain point threads extracted (11/19 via headless browser)
- 8 cross-cutting patient pain points identified
- 10 verbatim quotes captured
- 3 competitor deep-dives (Abridge, Viz.ai, Ceribell) — all 4 steps each
- Cross-competitor analysis with 8-pain-point gap matrix
- 5-layer AI taxonomy built (Layer 5 = pre-visit synthesis = the wedge)
- Layer 5 wedge thesis documented
- GastroNote references cleaned from all planning docs
- Cadence Notes / Points created for team syncs
Week 2 — Deeper Research + Doctor Interviews
In Progress
- Align on subspecialty — Headache/Migraine vs Cognitive/Dementia vs Epilepsy vs Movement Disorders
- Friend’s first doctor interview — validate the 8 pain points with a real neurologist
- Interview intake template — build the script/template for doctor interviews
- 13 quick-scan competitor teardowns — DeepScribe, Suki, Freed, RapidAI, Brainomix, Aidoc, NeuroQuant, Rune Labs, Empatica, Natus, encevis, Piramidal, DeepCura (~6h)
- Understand what GastroNote / other track actually built — need context, ideally a demo
- GPT-4 access confirmed? — clarify API keys before Week 3
Blocked / Waiting On
- Doctor interview scheduling (friend’s side)
- Subspecialty decision (blocking everything downstream)
Week 3 — Label Studio + First Training Experiment
Not Started
- Set up Label Studio — install, configure, import first dataset
- Download CHB-MIT EEG dataset — for seizure detection baseline
- Download ADNI MRI dataset — for dementia/cognitive baseline
- Pick subspecialty — blocks dataset choice
- Run baseline model — simple CNN/RNN on CHB-MIT or ResNet on ADNI
- **Friday Demo 1 — present initial training results to team
Week 4 — Market Data + Imaging AI Landscape
Not Started
- Market sizing — TAM/SAM/SOM for chosen subspecialty
- Doctor pain point survey — formal survey if interviews aren’t enough
- Imaging AI landscape — RapidAI, Brainomix, Aidoc, NeuroQuant teardowns
- **Friday Demo 2 — market data findings
Week 5-6 — Three Tracks (Individual Focus)
Not Started
Business Track
- Market opportunity report
- Revenue model analysis
- Pricing model for chosen wedge
Clinical Track
- Clinical workflow map for chosen subspecialty
- Subspecialty referral criteria
- Doctor interview synthesis
AI Track
- Prototype architecture (Layer 5: RAG + multi-agent LLM + caregiver channel + wearable)
- Dataset finalized
- Model architecture chosen
- First end-to-end demo
Week 7 — Deliverables Check
Not Started
- All 3 tracks cross-checked
- Gaps filled
- Week 8 presentation drafted
Week 8 — Presentation
Not Started
- Final deck built
- Thin-slice demo recorded or live
- Presentation rehearsed
Open Questions (Blocking Items)
| Question | Owner | Blocking | Priority |
|---|---|---|---|
| Which subspecialty? | User + Friend | Everything downstream | 🔴 Critical |
| What is GastroNote demo? | User | Context for differentiation | 🔴 Critical |
| Doctor interview script | Friend | Week 2 validation | 🟡 High |
| GPT-4 API access? | User | Week 3 training | 🟡 High |
| Deliverable = blueprint or working product? | Team | Scope for Weeks 5-8 | 🟡 High |
| Label Studio infra owner? | TBD | Week 3 setup | 🟡 High |
| Dataset access (IRB? real data?) | User | Week 3 training | 🟡 High |
| Friday Demo format? | Team | Weeks 3-5 | 🟡 High |
Running Notes
- Bonk is active development — separate from capstone. Do not mix.
- EchoDubBot / Oracle VPS — not relevant to capstone.
- Active model: MiniMax-M2.7 via minimax-oauth