VA
Visual notes

One idea per image.

Short, image-first engineering notes for quick review and sharing — closer to a technical cheat sheet than documentation. No article pages, no documentation tree.

KV cache and continuous batching — The KV cache trades memory for decoding speed; continuous batching keeps the GPU busy by admitting and retiring requests as sequences progress.

KV cache and continuous batching

GPU & Inference · inference · kv-cache · batching

Open image
Recall@K for retrieval systems — Recall@K asks whether the evidence you needed made it into the top K retrieved items before generation can hide the failure.

Recall@K for retrieval systems

Applied AI · rag · retrieval · evaluation

Open image
Retries and idempotency in agent tools — A retry is safe only when repeating an operation cannot accidentally duplicate the real-world side effect.

Retries and idempotency in agent tools

Applied AI · agents · reliability · idempotency

Open image