Empathetic AI that understands emotional context
AI USM is an AI company building empathy-centred artificial intelligence — emotion recognition from vision, voice and text, long-term contextual memory, and avatars that respond to how a person is, not only to what they typed.
What AI USM builds
AI USM is an artificial-intelligence company building empathy-centred AI: systems that read emotional context from vision, voice and language at the same time and adjust how they answer, not only what they answer. The stack is proprietary — multimodal emotion recognition, the AriGraph long-term memory layer and speech-synchronised avatars — and is not a wrapper around a third-party chat model.
The group operates as USM Tech Inc., a Delaware (USA) corporation, and AI USM LTD, registered in ADGM Abu Dhabi under number 28022 and endorsed by Hub71. The founder is Ildar Lukmanov.
- Multimodal emotion recognition across facial dynamics, prosody and text semantics
- AriGraph: graph-structured long-term memory of facts, events and emotional context
- Specialist avatars for health, fitness, nutrition and study support
- Five interface languages: English, Russian, Chinese, Czech and Arabic
Emotion recognition, not sentiment tagging
Sentiment analysis labels a sentence positive or negative. Emotion recognition in AI USM fuses three independent channels and resolves their disagreement: a smile that contradicts a tense voice does not silently win. On our internal validation sets the fused model reaches high accuracy, higher than any single channel on its own, with end-to-end inference under 200 ms so an avatar can change tone mid-conversation.
Where empathetic AI is applied
Healthcare is the demanding case: intake conversations that record how a patient reported a symptom alongside what they reported, and a multi-avatar consilium that shows disagreement between specialist perspectives instead of hiding it. Education and wellbeing use the same emotional signal to pace a session rather than push through disengagement.
AI USM assistants are decision support. They do not diagnose, they are not a medical device, and every health-adjacent surface states that.
- Healthcare AI: emotion-aware intake, triage support and patient follow-up
- Education: study companions that adapt pacing to engagement signals
- Human-AI interaction research: contextual, personalised response generation
Privacy and compliance by construction
Consent is explicit and per-modality: voice analysis can be allowed while the camera stays off. Raw video is processed for inference and not retained by default; what persists is the derived, anonymised state, which a user can inspect and delete. The platform is built against GDPR and HIPAA requirements and the DOH Abu Dhabi framework.