Work on merged models
We are a small team, and we are not hiring for listed roles right now. If you merge models — or want to — we would still like to hear from you.
Problems we live with
Fusion at inference
Lenses, merge prompts, shared-error hunting, audits, and the latency budget that keeps all of it usable.
Expert transplant
Moving expert layers between mixture-of-experts families through adapters that start at zero.
Weight merging
Combining fine-tunes of one base into a checkpoint that keeps what each one learned.
Evaluation
Measuring merged models honestly, including the prompts where merging makes things worse.
The platform
The router, the edge, accounts, billing and caching that put a merged model behind one API.
Writing it down
Docs and research notes that say exactly what we did and what we did not.
Small, careful, direct
We are based in Singapore and we are few. Everyone ships, everyone reads the logs, and everyone answers to the same rule: claim nothing you did not measure.
We care more about what you have built than where you studied. A merge you ran, a write-up of one that failed, or a careful evaluation of somebody else's is worth more to us than a CV.
What to send
A short note
Who you are and which of our tracks you want to work on.
Something you made
A repository, a model card, a paper or a post. Links are better than attachments.
One honest result
Something that did not work, and what you learned from it.