The Developer Blog

Every API call meets a total stranger — LLMs remember nothing. How AI memory systems actually work: short-term, long-term, and forgetting.

I almost paid for a dedicated vector database. Turns out Postgres does it. A practical pgvector tutorial with Supabase — setup to indexes.

Users searched "remove account," my docs said "delete account," search returned nothing. So I added AI semantic search to my Next.js site.

The AI never runs your functions — it just asks. A complete AI tool calling tutorial with real OpenAI and Claude API examples and code.

My first multi-agent system talked in circles and burned money. Here's the multi agent AI architecture that finally worked — and when to skip it.

I built an AI customer support agent that remembers users — short-term history, long-term facts, and knowing when to hand off to a human.

My AI API bill was growing faster than my users. Prompt caching cut input costs by ~80% — here's the exact restructure that did it.

Everyone says "train the AI on our docs." I tested RAG vs fine-tuning on a real project — here's which one to choose and when.

MCP finally made sense when I built a real server for it. Here's my Model Context Protocol tutorial — Claude talking to my own database.