AI
AI Assist make longer
A conversational assistant built on Next.js and modern LLM APIs — streaming responses, tool calls, and a chat surface tuned for real work.
Built around the Vercel AI SDK with streaming responses, tool calling, and a typed message protocol end-to-end. The hard parts were not the prompts — they were latency, partial state, and recovering cleanly from interrupted streams.
Stack: Next.js 14 App Router, TypeScript, Tailwind, Prisma + PostgreSQL for chat history, NextAuth for sessions, deployed on Vercel with edge runtime for the streaming routes.