Short, honest writeups on design decisions, mistakes, and reflections from building this portfolio.
Why I chose Server Actions over API Routes for credit operations in AuthorAI, and when I still use API Routes.
Implementing idempotent webhook processing to handle duplicate events and ensure credit balance accuracy.
Adding semantic search to AuthorAI using vector embeddings, and the tradeoffs involved.
How I introduced and fixed a race condition in the credit deduction system that allowed negative balances.
I initially built a complex cache invalidation system when simple TTL-based caching would have sufficed.
How I reduced AI API costs by 60% through caching, prompt optimization, and smart model selection.