Built university-industry deal platform in 14 days and reached top 5 at IDEASHACK
In just 14 days, this team vibe-coded CollabPilot—an AI co-pilot for university-industry commercialization deals—from user interviews to working MVP using Claude. The project earned a top 5 finalist spot at IDEASHACK 2026 in Warsaw.
Project Description
CollabPilot is an AI co-pilot that guides researchers, research institutions, and business partners through university-industry commercialization deals — from first contact to signed agreement. Vibe-coded in 14 days with Claude (chat) and Claude Design — from first user interviews to a working MVP prototype — for IDEASHACK 2026. https://www.ideas.edu.pl/en/wydarzenia/ideashack-2026/
Achievements
- 14 days, zero-to-MVP — vibe-coded end-to-end with Claude (chat) and Claude Design, starting from real user interviews and mentor validation, not a pre-set spec
- Top 5 finalist, IDEASHACK 2026 (of ~14–15 submitting teams) — live pitch + jury Q&A, Warsaw, June
- 8 design iterations (v1→v8) with versioned changelogs and audit trails, each tied to real user/mentor feedback
Tools
- Claude (chat) — user interview synthesis, ideation, journey mapping, personas, copywriting, presentation structuring
- Claude Design — vibe-coded the design system and high-fidelity MVP prototype
- Google Gemini - presentation & illustration
- Vercel — deployment
Methodology
1. Claude Project as system of record — all context (interviews, mentor notes, briefs, personas) centralized so every generation pulled from the same source.
2. Evidence-driven decisions — pivot to "post-match navigation" and rejection of a chatbot interface both came from mentor/interview input, not preference.
3. Full design chain run through AI — user research → product strategy → design strategy → UX → UI → usability testing, each stage feeding the next.
4. AI as co-pilot, not agent — every output shown with reasoning, human-reviewed before use.
Learnings
1. User research is crucial to get the right ideas for design — ungrounded prompts produce ungrounded output; AI is only as good as the spec it's fed.
2. Human taste and judgment play a vital role at every milestone — the work didn't disappear, it moved earlier: into the brief, the persona, the rejection criteria.
3. Humans own the quality of work, while AI is quick to implement — "vibe coding" without a spec is prompt roulette; with a spec, it's genuine leverage.
Relevant URLs
- Live demo: https://collab-copilot-7fxd.vercel.app/
- Video walkthrough: https://youtu.be/zDBKblb1KW4
Jingjing Yao
Written 10 August 2026 Submitted by the storyteller
Every story is written by a Mindvalley student or by someone our teachers have worked with. How we collect and check stories