Artificial intelligence
Bringing large language models (LLMs) into development work and into products: knowing how to talk to them, picking the right model and running it in the right place.
Writing precise, structured instructions: role, context, examples (few-shot), constraints and expected output format. Iterating and comparing results to get reliable, repeatable answers.
Understanding the strengths and limits of the major model families (OpenAI's GPT, Anthropic's Claude, Google's Gemini, Mistral, Meta's Llama): context window, reasoning, cost, latency, hallucinations. Picking the right model for the task.
Running open-source models locally (Llama, Mistral, Qwen…) to prototype with no API cost, work offline and keep sensitive data on your own machine.
Integrating models into Node.js and Next.js applications: streamed responses, structured JSON output, handling errors, rate limits and costs.
Using coding assistants and agents day to day, with method: context, plan, small iterations, review and tests. I describe the approach in a blog article.