RAG vs fine-tuning: how to choose
Both approaches improve LLM output quality. They solve different problems. Choosing the wrong one wastes months.
EmpiricaNotes
Notes
Short technical notes on AI evaluation, architecture, and production reliability.
The agent architecture decision
Single-agent or multi-agent? The answer depends on failure tolerance, observability, and whether subtasks are actually independent.
The real cost of LLM API calls
Token spend is the visible line item. It is rarely the largest cost. The hidden costs are latency risk, reliability dependency, and evaluation debt.
Evaluation before deployment
The most common pattern we see in production AI failures isn't a bad model — it's a good model evaluated on the wrong distribution.
Benchmark scores and production reliability
MMLU measures something real. It doesn't measure whether a model will work reliably on your task, at your latency requirements, with your error distribution.