LURA Research Core
RESEARCH LAB · TASHKENT · 2024

LURA AI.
The future of intelligence.
Our heritage.

Lura is an independent research lab building speech and language models for the Uzbek language — from real-time recognition to deep document understanding.

Explore projects Talk to the team
/ live products
02 shipped
/ language coverage
uz · ru · en + dialects
/ focus
STT · LLM document AI
/ 01 — Projects

Two products — one language stack.

We ship products ourselves to keep the whole chain — data, models, infrastructure — under one roof. That gives us speed and quality that are otherwise unreachable in the Uzbek market.
yozma.ai — Uzbek STT
/ in production

yozma.ai

Intelligent localisation and text editing. Speech-to-text for Uzbek with dialect awareness, both Cyrillic and Latin scripts. Real-time transcription, speaker diarisation, punctuation — infrastructure for media, contact centres and government.

WER 4.8% · RTF 0.18× · UZ-CYRL/LAT
Learn more
adolat.ai — document LLM
/ in production

adolat.ai

Automation and management of digital documents. An LLM service that translates entire documents — contracts, reports, legal and medical materials. Preserves layout, tables and signatures, and reasons over the whole file rather than isolated sentences.

UZ ⇄ RU · UZ ⇄ EN · LAYOUT-AWARE
Learn more
/ 02 — Research philosophy

Deep roots. High technology.

Lura is not just a company, it is a research lab whose goal is the harmonious fusion of Uzbekistan's rich cultural heritage with the frontier of artificial intelligence. We believe in AI that understands and respects the human being.
LURA research core
// LURA RESEARCH CORE
/ FIG.01 — LANGUAGE STACK
UZ.LM · v0.3 · LIVE
Uzbek manuscript and source code — Lura AI cultural gene in the algorithm
— 01
/ research

A cultural gene in the code

We engineer the future while respecting the roots. Every LURA algorithm is shaped by a deep understanding of Uzbek culture, language and history.

Human-centred Uzbek AI products — yozma.ai and adolat.ai design philosophy
— 02
/ product

Innovation with a soul

Building human-centred AI. yozma.ai and adolat.ai are only the beginning. We aim for systems that are intuitive and culturally adapted by default.

Lura AI engineers, linguists and historians collaborating on Uzbek language research
— 03
/ science bridge

A scientific bridge into the future

We unite the best AI engineers, linguists and historians. Together we build scientific bridges to preserve and grow the Uzbek language in the digital era.

/ 03 — Our technology

One language stack, assembled from six layers.

Every Lura product sits on the same core stack — from data collection to inference. It is an engineering choice: a shared base makes each new product cheaper and better.
// 01
Uzbek Speech Corpushundreds of hours of annotated audio, both scripts, dialects
/ data
// 02
Morphology-aware tokenizerhandles agglutination and the affix structure of Uzbek
/ pre-training
// 03
ASR encoder (yozma core)streaming speech recognition model in real time
/ speech
// 04
Document LLM (hujjat core)operates on the whole file: layout, tables, signatures
/ language
// 05
Domain adapterslegal, medical, financial — no retraining of the base
/ verticals
// 06
On-prem inferencebuilding closed-perimeter deployment — for sensitive data
/ deploy
LURA stack visualization
// CORE STACK
/ FIG.03 — STACK TOPOLOGY
06 LAYERS
/ 04 — Open lab

Uzbek AI is an open field. Build it with us.

We are looking for linguists, speech researchers and ML engineers who care about the Uzbek language — from corpora and morphology to speech models and document AI. Open to joint projects, residencies and methodology exchange.

Email the team
  • / 01Corpora & datasets
  • / 02Speech & ASR
  • / 03Morphology & NER
  • / 04Document understanding
  • / 05Low-resource models