Quartz 5

Cadence Notes - Points

3 min read

Cadence Notes / Points

capstone ai-internship neurology

Short doc for team syncs — questions to raise, decisions needed, problems faced + solutions. Updated after each cadence meeting.

Questions to Ask / Decisions Needed

  • Which subspecialty to focus on? (Headache/Migraine vs Cognitive/Dementia vs Epilepsy vs Movement Disorders) — friend is interviewing doctors, user is doing competitor research. Need to align on ONE before Week 3.
  • What exactly is GastroNote / what did the other track build? Context needed: demo? Product page? We need to know what “the parallel track” actually is before deciding how to differentiate.
  • What does “working product” mean for Week 8? Blueprint only? Thin-slice demo? MVP? The team needs to align on deliverable scope before Week 3.
  • Friday Demos — what’s the format? Who attends? How long? Are they internal or with doctors/friends?
  • GPT-4 access — do we have it? Week 1/2 research notes mention using GPT-4. Do we have API keys? Should we use Claude or MiniMax instead?
  • Doctor interviews — how many? Which specialties? Friend is doing interviews. What questions are we asking? Is there an intake template?
  • Label Studio — who sets it up? Week 3 mentions running the first training experiment. Who handles the ML infrastructure?
  • Data access — do we have real neurologist data? The plan mentions OpenNeuro, CHB-MIT EEG, ADNI. Are those accessible? Any IRB issues?
  • What is the “Big Vision” we commit to? The pre-visit synthesis layer is the technical wedge. But what’s the 1-sentence product vision for the Week 8 deck?

Problems Faced / Solutions

ProblemStatusSolution
Reddit scraping blocked by JS challengeSolved (partial)Headless browser via delegate_task subagent; 11/19 threads extracted
GastroNote references in planning docsSolvedPatched all 4 docs; original Planning & Questions.md left untouched (user-owned)
Abridge subagent didn’t save fileSolvedWrote the Abridge teardown directly using data from 4 web searches
Model switching mid-session (MiniMax / moa / minimax-oauth)OngoingActive model confirmed: MiniMax-M2.7 via minimax-oauth
Doctor interview intake templateNot startedNeed to build — suggested template in Week 1 research notes
13 quick-scan competitor teardownsNot started~6 hours of parallel subagent work; low priority until subspecialty chosen
CHB-MIT EEG dataset downloadNot startedNeeds to be set up before Week 3 training experiment
Label Studio setupNot startedNeeds ML infra owner before Week 3

Agenda Template for Weekly Sync

1. Progress since last sync (5 min)
   - What did we finish?
   - What didn't we finish?

2. Decisions needed today (10 min)
   - Subspecialty choice
   - Deliverable scope for Week 8
   - Friday Demo format

3. Blockers (5 min)
   - Data access?
   - API keys?
   - Doctor interviews scheduled?

4. Next week priorities (5 min)
   - Who owns what?

Open Threads (Low Priority but Worth Revisiting)

  • Bonk is still in active development — Bonk engineering work is separate from this capstone. Keep separate.
  • EchoDubBot / Oracle VPS — not relevant to capstone.
  • Memory consolidation reminder — MEMORY.md is at 94% capacity. After Week 2, consolidate to make room for new capstone learnings.

Graph View

Backlinks