Four directions where AI
still has room to grow up.
Manas AI's research agenda is organised around four interlocking thrusts. Each is animated by the conviction that the next decade of AI belongs to systems that are physically grounded, phenomenologically informed, and ethically deployed.
Cutting-edge AI applications
We pursue applied AI where the science is hard and the stakes are real — from indigenous LLM tooling and retrieval over classical knowledge corpora, to multimodal inference for healthcare, education, and field instrumentation.
- Wisdom-RAG: retrieval over Indic philosophical and scientific texts.
- Indigenous LLMs for Indian languages and discourse styles.
- Multimodal pipelines for affect, attention, and arousal.
- Deployment patterns optimised for low-bandwidth, low-power settings.
Physics-informed AI for physical AI
Generic deep learning ignores what physics already knows: conservation, symmetry, scaling, causality. We build hybrid architectures — neural-symbolic, equation-embedded, physics-residual — for systems that must reason about the material world, not just match its surface statistics.
- Physics-informed neural networks for signal recovery and inverse problems.
- Equation-aware models for biosignals (HRV, respiration, EDA).
- Symmetry-respecting architectures for sensor fusion.
- Embodied agents whose actions respect physical priors.
Consciousness-based research
What does it mean to model attention without modelling the attender? We treat consciousness studies — both contemporary and classical — as first-class research, drawing especially on the phenomenological precision of Indic traditions of inquiry into mind.
- Computational models of attentional stability and reflective awareness.
- Bridging first-person methods with third-person measurement.
- Comparative inquiry across cognitive science and contemplative philosophy.
- Ethical frameworks for AI that interacts with subjective experience.
HCI advances for healthcare and wellness
The most powerful AI in the world is useless if it cannot reach the people who need it. We design low-cost, low-power, low-friction devices and interfaces — instruments a community health worker can carry, a meditator can wear unobtrusively, and a school can afford for every classroom.
- Non-intrusive sensing — no headgear, no skin compression.
- Offline-first interaction patterns for low-bandwidth environments.
- Multilingual, culturally legible UI for diverse Indian contexts.
- Open hardware specifications wherever feasible.
See the platform that ties it all together.
Our four research thrusts converge in a single platform — sensing, inference, guidance, and learning.
Explore the platform