Sovereign Foundation Models, Speech Corpora & Open Datasets
Showcase of open research projects and sovereign technologies developed by ISSAI (Nazarbayev University) and the Kazakhstan AI engineering community.
38+ Projects
1,200+ Hours
70B Params
100% Open Access
Flagship 70-billion parameter sovereign language model of Kazakhstan trained on Kazakh, Russian, and English corpora.
Compact and fast KazLLM 8B model for local edge inference, enterprise RAG, and conversational systems.
Neural machine translation model supporting Kazakh, Russian, and English with state-of-the-art syntactic accuracy.
Qolda multimodal vision-language model understanding images, documents, and visual questions in Kazakh.
Open community language model fine-tuned for Kazakh text generation with 4-bit and 8-bit GGUF quantization.
Instruction-tuned Kazakh language model for summarization, reasoning, and automated text generation.
Family of ultra-lightweight SLMs (270M–1.7B) optimized for on-device mobile execution and embedded IoT.
Generative vision model trained on Central Asian cultural heritage, traditional ornaments, and historical scenes.
Pre-trained RoBERTa model for the Kazakh language, optimized for text classification, named entity recognition, and semantic embeddings.
Bilingual Text-to-Text Transfer Transformer model for abstractive summarization, rewriting, and machine translation in Kazakh.
State-of-the-art automatic speech recognition for 8 Turkic languages with robust dialect handling.
The largest open Kazakh speech corpus: over 1,200 hours of high-quality transcribed audio across diverse regional accents.
Neural text-to-speech synthesis system for Kazakh featuring studio-grade male and female voices.
First emotional speech synthesis system for Kazakh with control over anger, joy, sadness, surprise, and neutrality.
Conversational ASR engine power of the SAIDA voice assistant robust to acoustic background noise.
Lightweight ultra-fast neural speech generator for Kazakh based on Kokoro architecture with low latency.
Fine-tuned Whisper Large v3 Turbo on Kazakh speech with GGUF quantization for real-time transcription.
Self-supervised acoustic foundation model for Kazakh phoneme recognition and speech feature extraction.
Dataset of short Kazakh voice commands ('open', 'close', 'stop', etc.) for smart devices and robotics.
Cross-lingual speech recognition system with strong support for Kazakh-Russian code-switching.
The standard Named Entity Recognition benchmark for Kazakh with over 112,000 annotated sentences across 25 categories.
Kazakh Question Answering Dataset in SQuAD format for evaluating machine reading comprehension in Kazakh.
AI2 Reasoning Challenge localized for Kazakh and Russian to test complex scientific reasoning in LLMs.
Massive Multitask Language Understanding (MMLU) benchmark in Kazakh covering 57 diverse academic domains.
Complex instruction following benchmark evaluating language models on verifiable structural constraints in Kazakh.
Multi-step mathematical reasoning benchmark translated and culturally adapted into Kazakh and Russian.
Curated parallel corpus for sentence alignment, bilingual retrieval, and translation model training.
Standardized evaluation toolkit for running reproducible benchmark tests on Kazakh NLP and vision models.
Open structured legal corpus of Kazakhstan legislation, codes, and regulatory acts for LegalTech AI.
Adversarial prompt dataset for red-teaming, jailbreak testing, and safety alignment of Kazakh language models.
Aligned parallel bilingual corpus of over 1.2 million Kazakh-English sentence pairs curated for neural machine translation.
Groundbreaking multimodal dataset combining synchronized thermal and visible video streams paired with high-quality audio.
Thermal facial landmark detection and tracking capable of robust operation in total darkness and varying temperatures.
Human pose estimation in the thermal spectrum for healthcare monitoring, search-and-rescue, and night vision.
Privacy-preserving human action recognition on thermal video streams for industrial safety and smart cities.
Neuromorphic event-based camera dataset for high-speed microsecond face detection and landmarking.
Deep learning thoracic radiograph diagnostic module assisting doctors in detecting pulmonary anomalies.
Photorealistic talking-head digital human generator with synchronized lip-motion for Kazakh speech.
Visual dataset of Central Asian cuisines for food segmentation, volume estimation, and nutritional assessment.
Video corpus and 3D pose landmark dataset for automatic Kazakh sign language recognition and inclusive accessibility tools.
Detection and OCR model customized for Kazakhstan vehicle license plate formats and regional font geometries in Smart Cities.