After three hackathons (top 3, top 5, DNF), the difference between "we built something" and "we finished and placed" was never raw skill. It was structure. This is the playbook we now run on day one.
The Demo Sentence (Before Hour 3)
Every successful project we've built starts with one sentence:
"When [specific pain] happens, [product] does [specific thing] in [specific time/effort]."
Examples:
- "When your service breaks at 2 AM, SentryAI tells you what changed and what to do first."
- "When a non-technical teammate needs an SQL query, QueryBuddy turns their question into a safe, reviewed query."
If you can't write the sentence, you don't have a project yet โ you have a vibe. The sentence is the scope. Anything that doesn't serve the sentence gets cut later without a fight.
The First 3 Hours Decide Everything
Hour 0โ1 Pick the problem. 5 candidates max, score each:
(1) pain is real? (2) can we demo it? (3) judges will get it in 30s?
Kill the losers. No sunk cost.
Hour 1โ2 Architecture on paper. What is the ONE core flow?
What are we deliberately NOT building? Write the cut list.
Hour 2โ3 Spikes: prove the riskiest part WORKS before building around it.
(LLM call? DB? Auth?) If the spike fails, pivot NOW โ cheap now.
[!WARNING]
- The hour-2 mistake that sank our DNF: we spent 4 hours on a data pipeline that "might" be needed. The actual product needed no pipeline. Spike the risk, don't build the foundation.
Team Roles (3โ4 People)
| Role | Owns | Anti-pattern |
|---|
| Builder (backend) | Core flow, API, data | Building 5 microservices |
| Builder (frontend) | The demo surface | Designing for a product, not a demo |
| Builder (AI/integration) | The novel part | Chasing model perfection |
| Point person | README, demo script, judge comms | "We'll do it at the end" (there is no end) |
The point person role is non-negotiable. The README, the demo rehearsal, the pitch โ these are 40% of the score and they don't exist unless someone owns them from hour 10.
The Cut List (Hour 20 Ritual)
At hour 20, everyone writes their top 3 "nice features" on a board. Then:
For each: does it make the demo sentence MORE true?
Yes โ keep, schedule for last 4 hours ONLY if core is stable
No โ cut, publicly, and stop talking about it
Features we've cut at this ritual that we're still glad about: dashboards, authentication, mobile, analytics, settings pages. The demo IS the product at this stage.
Demo Rehearsal (Hour 36+)
The demo is a production. We rehearse with the exact rules:
- 90 seconds, hard limit. Time it. Cut until it fits.
- Run it live from a clean state โ no pre-clicked buttons, no cached data the judge can't see
- One fallback: if the live demo dies, screen-recorded video from hour 30 plays. It has never been needed, and it has never been unneeded.
- The script has one line of humility: "Here's the one thing we didn't have time for, and here's why we chose not to build it." Judges respect scoping.
The Mistakes That Cost Us Placement
- The DNF โ built a pipeline that wasn't the product. Lesson: spike first.
- Top 5 (not top 3) โ great build, mediocre README written in the final 2 hours. Lesson: the point person owns it from hour 10.
- The 25-minute demo โ we kept adding "quick" explanations. The judges' attention is the scarcest resource in the room. 90 seconds.
The Winning-Strategy Short List
- Narrow beats broad. "AI log diagnosis for on-call" beat "AI for developers" โ specificity is credibility.
- Real pain beats novel tech. Judges have seen 14 LLM apps; they haven't seen a good answer to a problem they have.
- Boring infra, interesting product. Managed DB, free API tiers, no custom k8s. Your 48 hours are for the idea.
- Honest limitations in the README signal engineering maturity. "We know this is a demo-grade ingest" scored us credibility points.
- Sleep is a feature. A 90-minute rotation kept us sharp for the pitch. Tired teams demo like it.
The Hour-by-Hour Skeleton
00โ03 Problem, demo sentence, architecture, spikes
03โ12 Core flow end-to-end (ugly is fine)
12โ20 Polish the ONE flow, start README
20 THE CUT RITUAL
20โ32 Build the cut scope, demo script, record fallback video
32โ40 Rehearse 10ร, fix demo bugs, final README pass
40โ46 Sleep. Seriously.
46โ48 Final rehearsal, pitch practice, submit early (hour 47)
A hackathon is not a test of what you can build in 48 hours. It's a test of what you can finish, demo, and explain in 48 hours. Build accordingly.
Discussion (0)
Sign in or create an account to join the discussion.
No comments yet. Start the discussion.