We run an applied AI lab that works on hard problems — from custom model training to complex enterprise deployments — while publishing research that advances the field.
We work at the intersection of Physical AI, low-resource language AI, and document understanding. We publish, open-source, and build things that don't exist yet.
100+ projects across enterprise and growth-stage companies. We handle complexity — from custom model training to full-stack Voice AI pipelines and Physical AI systems.
From custom model training to full production deployments — across every AI modality.
We believe the next frontier of AI is understanding and acting in the physical world. Our research targets two hard problems: training VLMs to reason spatially and embody real-world physics, and training JEPA-based models to plan over long horizons.
On the applied side, we're already deploying this — anomaly detection, factory monitoring, defect detection, traffic intelligence — and plan to evolve the lab toward a Physical AI company.
See our Physical AI research →Joint Embedding Predictive Architectures trained for physical world planning, enabling robots to reason over extended action sequences.
How to adapt SOTA text embeddings to low-resource languages using only 10k noisy translations. Applied to Armenian — methodology generalises broadly.
ColPali-style multimodal embeddings for visual document retrieval, ranked #1 on ViDoRe globally. Open on Hugging Face.
Full voice pipeline — ASR, LLM, TTS — with fine-tuned models for Armenian, English and Russian. Sub-300ms latency, 98.7% accuracy.
Industrial anomaly detection deployed across 12 production lines. 40% reduction in defect pass-through.
Multimodal RAG pipeline over complex legal documents. Fine-tuned for jurisdiction-specific reasoning.
We respond to every serious inquiry — usually within a day.
We'll be in touch soon.