Transcript-free persistent memory in event-stream models: the PNS-World benchmark and its causal evaluation.
NURGALI Kadyrbek PRO
nur-dev
AI & ML interests
None yet
Recent Activity
published a dataset 1 day ago
nur-dev/farabi-knowledge-base updated a dataset 1 day ago
nur-dev/farabi-knowledge-base updated a model 9 days ago
nur-dev/farabi-speech-modelsOrganizations
Multilingual speech, multimodal perception, language-guided robotics, and compact embodied-AI models, datasets, and reproducibility artifacts.
-
nur-dev/multilingual-robot-grounding-causal-audit
Updated • 156 -
nur-dev/farabi-0.6B-agent-rag
Text Generation • 0.6B • Updated • 10.9k • 1 -
nur-dev/Qwen3-ASR-0.6B-kk-ru-en
Automatic Speech Recognition • 0.8B • Updated • 3 • 3 -
nur-dev/ait-syn-0.6b-trilingual-v30
Text-to-Speech • 0.9B • Updated • 14 • 5
Gated Kazakh/Russian/English datasets for retrieval-augmented and tool-using LMs, built by controlled augmentation (verifiable-by-construction).
Research artifacts connecting PAT-ER typed event-role side state with STRATA head-level typed predicate-graph substitution.
Evidence registry, derived release metadata, and manuscript code for the frozen CAMP-ER Episode-Core R1A early sepsis warning study.
Frozen-frozen speech-LLM adapters + counterfactual training. Code: github.com/Nurgali-Kadyrbek/frozen-speech-llm-stress
Transcript-free persistent memory in event-stream models: the PNS-World benchmark and its causal evaluation.
Research artifacts connecting PAT-ER typed event-role side state with STRATA head-level typed predicate-graph substitution.
Multilingual speech, multimodal perception, language-guided robotics, and compact embodied-AI models, datasets, and reproducibility artifacts.
-
nur-dev/multilingual-robot-grounding-causal-audit
Updated • 156 -
nur-dev/farabi-0.6B-agent-rag
Text Generation • 0.6B • Updated • 10.9k • 1 -
nur-dev/Qwen3-ASR-0.6B-kk-ru-en
Automatic Speech Recognition • 0.8B • Updated • 3 • 3 -
nur-dev/ait-syn-0.6b-trilingual-v30
Text-to-Speech • 0.9B • Updated • 14 • 5
Evidence registry, derived release metadata, and manuscript code for the frozen CAMP-ER Episode-Core R1A early sepsis warning study.
Gated Kazakh/Russian/English datasets for retrieval-augmented and tool-using LMs, built by controlled augmentation (verifiable-by-construction).
Frozen-frozen speech-LLM adapters + counterfactual training. Code: github.com/Nurgali-Kadyrbek/frozen-speech-llm-stress