290 AI Iterations: No-Code Full-Stack App in 7 Days
Non-engineer built Where2Eat group dining app in 7 days using v0, Claude, GPT after 289 failures. Key: Feed v0 code to Claude for optimized prompts, cutting costs 70% and fixing circular bugs. Reduces group decisions from 47 messages/3 hours to 10 minutes.
Fix Group Dining Bottleneck by Centralizing Shortlist & Align
Group dinners fail at stage 3 (Shortlist & Align) of the 5-stage process—taking 1+ hour vs. 15min scheduling, 20min preferences, 30min consensus, 5min confirmation. Hosts waste hours cross-referencing dietary needs, budgets across Google Maps/Yelp; chats bury links under memes; only 30% participate, leading to safe chains like Chipotle.
Where2Eat cuts this to <10min for hosts, <2min for participants: Host creates event/link (30s), monitors real-time dashboard, gets AI-curated top 10 restaurants scored on cuisine match, dietary compatibility, budget fit, distance via Google Places API (e.g., 'Japanese' expands to sushi/ramen/izakaya). Participants answer 4 questions (cuisine, dietary, budget, distance), vote on visual cards with photos/ratings/group fit %. Results show winner with booking links.
Metrics: Host time (hours→minutes), participation (30%→100%), satisfaction (delightful spots vs. mediocre). Beats swipe apps (no filters) and OpenTable (corporate-only) by organizing info first, enabling consensus.
Roadmap adds user suggestions, direct Resy/OpenTable booking. Monetize via affiliate commissions per reservation, then white-label for platforms—diner social layer boosts group bookings/network effects.
Validate Ideas and Build MVP with AI Tool Shootout
Score 5 ideas in GPT by build feasibility (1 week), market gap, free data access—Where2Eat topped. Research competitors/revenue via GPT Deep Research/Pitchbook.
Phases: Day 1 validation; Days 2-3 PRD (GPT/Claude critique), FigJam/Figma wireframes (AI suggestions inspired manual tweaks); Days 4-6 dev; Day 7 polish.
Shootout same PRD/wireframe/prompt: Replit best prototype but credit limits; Bolt pop-ups distract; Lovable Figma import too complex. Vercel v0 won: $50 cap, 3-click Figma connect, version slider for 290 iterations, 'ding' notification. Backend: Supabase. Use Google Cloud free tier ($300).
Polish: Descript for one-take demos (edits 'ums'/accents via text); Willow condenses prompts from novels to haikus.
Master Iterations: Claude-Powered Prompt Engineering Saves 70%
Early: Direct v0 fixes hemorrhaged credits, created circles (1 fix breaks 2). Pivot at iter 30: Copy v0 code to Claude—'Here's what I want, tried, write v0 prompt'—dropped costs 70%, fixed root issues. Non-engineer ships full-stack (frontend/backend/DB/API) in 7 days, proves problem-solving for job hunt.
Trade-offs: AI prototypes fast but need human validation; v0 editable code key vs. black-box tools. Outcome: From endless chats to one link, turning group pain into restaurant revenue.