befound.labsWhatsApp
HomeSample designsThe labEngineeringWritingPricing

Writing

Notes from things that went wrong.

Not tutorials. Each of these started as a bug I had to sit with — a memory profile that made no sense, a retrieval system that was confidently wrong, an agent graph with three boxes that didn't need to exist.

Latest·Python·

The string that ate our memory

A streaming endpoint was building its response one token at a time with s = s + chunk. It worked. It also allocated about twenty gigabytes to produce seven hundred kilobytes.

Read it — 6 min

Everything else

Agents

5 min

When a second agent earns its keep

Most multi-agent systems are one agent wearing a hat and paying for two. A test for whether the second one is real, and what to do instead when it isn't.

Retrieval

5 min

Your RAG isn't broken. Your chunks are.

Almost every retrieval system I've been asked to fix was failing before the model was ever involved. If the right passage isn't in the context, no prompt saves you.

Web

5 min

Two seconds is a budget, not a wish

Why most small business websites are slow, why the usual advice doesn't fix it, and how to spend a page-load budget on purpose.

These get cross-posted to Medium. If something here is wrong or you've hit the same thing from a different angle, I'd like to hear about it — contact@befoundlabs.in.