Insights

Notes from the build

Writeups from work we've actually done — the decisions, the things that broke, and the numbers that came out. No trend pieces.

14 July 20266 min read

Most RAG systems fail at retrieval, not generation

When a retrieval-augmented system gives a bad answer, the instinct is to change the prompt or the model. Usually the right passage was never in the context.

RAG
LLM systems
Evaluation

Read the note

9 June 20267 min read

Why we split ML inference into three services

Ingest, inference, and delivery fail differently and scale differently. Putting them in one process means the slowest one sets the rules.

System design
Kubernetes
Kafka

Read the note

18 May 20266 min read

Cutting a 600 MB model app down to 70 MB

ResoNet needed to run respiratory screening entirely on-device, in places with no reliable network. The model was never the problem — the runtime was.

Edge AI
ONNX
Mobile

Read the note

Have a problem worth solving with software?

Tell us what you're building. We'll help you scope it, build it, and ship it.

Get in touch