| name | hackathon-brainstorm |
| description | Brainstorm, evaluate, and blueprint award-winning hackathon projects using a bumwad-coding (architecture-phase) methodology with mandatory data/API feasibility scouting. Use whenever the user mentions a hackathon, hackathon ideas, hackathon prep, sponsor tech challenges, demo day projects, prize categories, or wants to pressure-test a hackathon concept — even if they don't say "brainstorm." Pushes past generic ideas toward projects judges remember and that actually ship. |
Hackathon Brainstorm (Bumwad v2)
5 phases: Research → Concept → Data & API Scouting (mandatory) → Design Development → Refinement.
Core winning criteria (required)
- Non-obvious sponsor tech use — sponsor would screenshot it. Heuristic: use their tool on a corpus they'd never think to index.
- <60s wow moment — name it explicitly or concept fails.
- Narrative hook — one sentence on stage/in video.
Amplifiers (never bolt on)
Cultural authenticity · Real stakes · Domain expertise
Phase 1 — Research
Ask in one batch: hackathon, deadline, submission format (live vs async video), theme, sponsors, every prize, rubric, team, unfair-advantage angle (required), constraints. Then web_search the hackathon page and past winners.
Phase 1.5 — Prize Structure Analysis (always)
Compute (social+virality points) vs (1st−3rd delta). If virality ≥ delta → distribution-primary hackathon, add Shootability ×2 and Audience leverage ×1 to Phase 2 rubric.
Phase 2 — Concept (8–10 ideas, ranked)
| Criterion | Weight |
|---|
| Non-obvious sponsor tech | ×3 |
| Wow moment clarity | ×3 |
| Narrative strength | ×2 |
| Cultural authenticity | ×1 |
| Real stakes | ×1 |
| Domain expertise | ×1 |
| Data feasibility | ×2 |
| Build feasibility | ×2 |
| Prize fit (name it) | ×2 |
Data and build feasibility are separate scores. A 5-on-build / 2-on-data concept is a trap. Flag top 3.
Phase 2.5 — Data & API Scouting (MANDATORY, web_search every claim)
Data table per top-3 concept:
| Need | Source | Access | License | Size | Effort | Fallback |
|---|
API table:
| Capability | Sponsor API | Backup | Rate limits | Auth | Demo cost |
|---|
Go/no-go: any row with High effort + no fallback → concept drops or gets scope-cut now.
curl-before-commit: prompt the builder to hit each sponsor API with curl before Phase 3.
Phase 3 — Design Development
- Logline, 60s demo script (timed beats), Narrative hook, Sponsor tech map
- Technical architecture (required):
- Data pipeline — source → cleaning → chunking (token size) → embedding model (named, dims) → vector store (named, index config)
- Retrieval path — query → top-k → re-rank → prompt template (actual text)
- Generation path — API, endpoint, params, cached vs live
- Frontend — framework, hosting, state
- Real vs. faked-for-demo table (every demo fakes something — make it explicit)
- Secrets & rate-limit math
- Deploy target (named)
- Scope cuts · Risk log (top 3) · Prize target (compute floor for virality-primary)
Phase 4 — Refinement (red team)
- 20-min-later memory test · laziest competing version · wow-fail backup · which sponsor champions internally · shareable permalink artifact?
Output format
Markdown, ranked tables, filled scouting tables, skimmable on phone at 2am.
Anti-patterns
- "AI-powered [thing]" with no twist · chatbots as primary UI · dashboards without stories · 10-min explanations · sponsor tech as plumbing
- Corpora that don't exist yet + >20% of time budget to build
- APIs never hit with curl before concept-doc written
- Live backends during async video recording (cache the hero output)
Reference
See references/past-winners.md for encoded heuristics (physical-object-on-camera, unfair-advantage corpus, sleeper-category math, cache-the-wow rule).
