Engineering·6 min read
Your search API isn't the expensive part
Teams optimize the wrong line item. A breakdown of where the money actually goes when an agent reads the web.
Read post →Blog
Notes from building retrieval infrastructure.
What we're learning about search, grounding, and the economics of feeding models the web.
Research·8 min read
Grounding is a retrieval problem before it's a prompting problem
Most hallucination in RAG systems traces back to thin or contradictory sources, not to the model. What to do about it.
Read post →Engineering·5 min read
Designing an agent around a one-second latency budget
Interactive agents have less time than you think. How to spend it across retrieval, reasoning, and generation.
Read post →