real software, beautifully made.

Kanata is a software studio from India. We design and build apps and AI products, from the first screen to the servers underneath.

Latest release: BeatMind, an AI music workspace, built and shipped in 24 days.

how it looks.

how it works.

what runs underneath.

three builds, measured.

Real products, with the numbers we measured and the mistakes we caught.

BeatMind

Client work · AI music workspace · 2026 · Live

Drop in a song and get back five stems, with every section and chord mapped, in an editor you can work in.

  • 70 s → 23.4 sto separate a four-minute track, once we found the GPU sitting idle
  • 24 daysto build and release it
  • 93+people finished a track and rated it 4.5 out of 5 in the first days after launch (self-reported)
read the build
Placeholder for a screenshot of the BeatMind editor: five stems down the side and the song’s sections across the top.

Vivid

Client work · Script to storyboard · 2026 · Built

Write a scene and get a storyboard of cinematic pictures. We rebuilt the engine that makes them, and built the test that says whether any change makes them better or worse.

  • 42 picturesin a fixed test of 12 stories, run the same way every time
  • 5.9×less GPU work per picture from a turbo setting, which we did not ship because it made faces worse
  • 6× cheaperto run a full test, from about ₹250 to about ₹42
read the build
Placeholder for screenshots of Vivid: a storyboard of frames from one story.

The airport parking app

Client work · Mobile app, Android and iPhone · 2026 · In development

Find parking, book a space and pay in the app, with your number plate as the pass. Working on Android phones today, not on the stores yet.

  • 1,900automated tests, all passing at the last full check on 9 October 2026
  • 1 codebasefor Android and iPhone
read the build
Placeholder for the first screen: the map of nearby car parks.
Placeholder for the second screen: booking a space.
Placeholder for the third screen: the pass, which is your number plate.

anything software, top to bottom.

believed, measured, changed.

We write down what we got wrong. A case study without its corrections reads like marketing.

we believedwe measuredso we
A bigger GPU would make generation faster.Diffusion was 0.42 s of a 16.32 s run. An L40S beat both an A100 and an H100 on speed and on cost.Did not upgrade. The cost that mattered was the idle time after each run: six times the generation itself.
Separation was running on the GPU.A missing library had pushed it onto the CPU while the GPU sat idle.Fixed it: 70 s down to 23.4 s on a four-minute track, re-checked on two kinds of GPU.
A setting that cut GPU work 5.9× was a clear win.Faces got worse and people were duplicated.Did not ship it. Kept the slower setting.
A checkbox is enough to prove who owns a track.A checkbox proves nothing about where audio came from.Deleted it. An export now has to pass a fingerprint check.

the interface, and everything under it.

Kanata is a small software studio from India, run by Parth Tiwari. He designs the interface, builds the backend and puts the AI to work, so what you get is one piece of work and not three handoffs.

The studio started in September 2026, a few weeks after Parth released BeatMind, an AI music workspace he built alone in 24 days. That build showed what one person who designs and engineers can take from idea to a live product. Kanata is that way of working, now open to other people’s products.

We measure before we believe. When a measurement shows we were wrong, we write it down and change course. The record above is that list.

what are you building?

Two lines is enough. Parth reads every message himself and replies from his own inbox.

i want to
rough budget (optional)
Or write to parth@kanata.studio

Sent.

Menu