Case Study

18 days

Solo Built

Kanbi Board

You take notes everywhere, meetings, calls, random ideas, but none of it becomes action. You spend 20 minutes daily manually converting notes into tasks. Context gets lost. Priorities get buried. Your productivity leaks through the cracks between tools.

1. The Founder's Problem

2. Why Traditional Solutions Failed

✗Note apps don’t do tasks. Task apps don’t do notes. You’re stuck toggling between both.
✗Copy-paste loses formatting, links, and context. You end up with flat, meaningless task titles.
✗No AI reliably understands unstructured notes. Most tools choke on anything that isn’t a checklist.

3. What We Built

✓AI task extraction with dual-provider failover (Gemini + Groq) for 95%+ accuracy on messy input
✓Drag-and-drop Kanban board with optimistic UI updates for instant feedback
✓Supabase Row Level Security ensuring each user sees only their data
✓Stripe subscription integration (test mode) to prove production-grade payment flows
✓JSON export with proper data serialization for backup and portability

4. The Process

WEEK 1

Database schema (3 days), Supabase RLS policies (2 days), Auth flow (2 days)

WEEK 2

AI integration + prompt engineering (4 days), Kanban UI (3 days)

WEEK 3

Stripe webhooks (2 days), Testing edge cases (2 days), Deploy (1 day)

5. Results & Metrics

Users

23 signups

Revenue

$0 (test mode)

GitHub

11 stars

Performance

Task extraction in <3 seconds, AI failover in <500ms

Code

95.4% TypeScript, 19 commits

6. Tech Stack & Why

Next.js 16App Router for clean server components. Faster TTFB than Pages Router.

SupabaseFree PostgreSQL with RLS. Cheaper than Firebase and handles relational data properly.

Groq + GeminiGroq for speed (800 t/s), Gemini for accuracy. Dual failover means zero downtime on AI.

StripeGold standard for SaaS billing. Better docs and SDK than any alternative.

40% of the build time went into prompt engineering, not code. Consistency is harder than speed.
RLS policies look simple on paper. They break in production if you only test with one user.
Stripe webhooks will fire duplicates. Idempotency keys are non-negotiable. Learned this with a double-charge scare.

7. Lessons Learned

40% of the build time went into prompt engineering, not code. Consistency is harder than speed.
RLS policies look simple on paper. They break in production if you only test with one user.
Stripe webhooks will fire duplicates. Idempotency keys are non-negotiable. Learned this with a double-charge scare.

What This Actually Proves

✓I ship fast: 7 MVPs in 14 to 21 days each. Not prototypes, actual working products with auth, payments, databases.
✓I'm honest about metrics: 10 to 50 users, $0 revenue (test mode), 7 to 10 GitHub stars. No fake numbers.
✓I know what to cut: Real-time features, team collaboration, mobile apps. Ship core value first.
✓I learn from mistakes: RLS policies are hard. Prompt engineering takes 15+ iterations. Financial calculations need edge case testing.
✓I justify tech choices: Supabase over Firebase (cheaper, better for relational data). Groq over OpenAI (10x cheaper, 800 tokens/s).

More case studies

Built in 18 days

Built in 20 days

Built in 21 days

Built in 19 days

All Projects