A small company with an unfashionable bet.
Most AI products pick a model provider and build on top. That is a reasonable bet for a quarter and a bad one for a decade: the frontier moves, prices halve, and the model that was obviously best in January is merely fine by June.
MuseMesh assumes the opposite — that models are becoming interchangeable parts, and that the durable work is the layer that makes swapping one for another boring. One key, one bill, one request shape, and honest information about what each provider does with your data.
Mume AI is that layer with a face on it, and it is also how we find out what is missing. Every capability on the platform was built because the app needed it first.
The founder
MuseMesh is Kushal Sharma — a decade of building at scale before this, and not in one lane. Six years at Flipkart across Android, iOS and backend: two open-source libraries, the app cut from 50MB to 10MB, a widget framework rewritten to render at 60FPS, and a review backend scaled tenfold for sale events.
Machine learning there too, years before the LLM era — transfer learning and image classification in production, with its own prediction service, which took image moderation from days to minutes. Then zero-to-one at Virgio, where an e-commerce platform reached 100k daily users inside a year and product teams could change the UI without shipping a release.
This is his second company. Alphawolf came first — two Android apps he built and published alone, past half a million users. He maintains open-source agent tooling alongside Mume AI, and still designs it, writes it, and answers the support email.
The company
Working on something adjacent?
We are easy to reach and quick to reply.