Naamche LabsNAAMCHE LABSनाम्चेJoin Us

AI researchfrom the Himalayas.Built for Nepal. For the world.

We build speech models, language systems, and agents for Nepal, released openly with the evidence to back them up.

Building the intelligence layer
for Nepal's digital future.

Frontier Research

Long-horizon work on language, agents, and efficient models, not quick demos.

Nepali First

Systems designed around how Nepali is actually spoken, written, and used.

Open and Responsible

Open weights, open benchmarks, and honest limitations with every release.

why we exist

at this stage, almost everyone agrees that ai is one of the biggest technological revolutions of our time, potentially even bigger than the internet. it is becoming something as fundamental as electricity.

most of the major innovation in ai is currently coming from western labs, where english is the primary language. as nepali people, we are often a minority in this revolution. nepal and the nepali people have had a meaningful impact on the history of the world, and we don't want that history, culture, and identity to be overlooked or forgotten in this new technological era.

we don't want to simply be observers.

introducing

naamche labs.

we are a small group of people doing everything we can to contribute to this revolution and create a meaningful impact. despite having limited resources and facing various constraints, we are trying our best to build something meaningful for nepali people and contribute to the future of ai.

Kriti model family

Two open models.
One for every pipeline.

Kriti V1 is our wideband Nepali ASR model — rank #1 of 19 systems, 24.08% WER, 630 hours of training audio. Kriti Telephony adapts it for phone calls and call-centers, reaching new state of the art on the NepTel benchmark.

Kriti V1 WER24.08%
Telephony WER32.38
Parameters119M
Benchmark rank#1 of 19
Kriti V1
Open weights, MIT

Nepali speech recognition

Nepali, heard clearly.

Kriti is a 119M-parameter Nepali-only RNN-Transducer. It shares rank one on the development benchmark and records the lowest raw word error rate in the evaluated field of 19 systems.

Live parameters
119M
Primary WER
24.08%
Benchmark rank
#1
Training audio
630 hrs

Serious AI research.
Built from Nepal.

We are a small team in Kathmandu. We are not hiring for set roles right now, but we always want to meet people who could shape the lab.

Follow the work.

A short email when we publish a model, a benchmark, or a research note. No marketing.