Research

Intelligence that can
show its work.

One principle runs through everything we study: a system that cannot explain itself cannot be trusted with a decision that matters. In robotics that means interpretable policies. In education it means a grade a teacher can interrogate.

Neural-symbolic policy learning

Active

Deep RL for adaptability, symbolic logic for safety and interpretability. A policy you cannot explain is a policy you cannot deploy.

Physical AI
Interpretability
Robotics
Causal inference

Assessment AI and explainable grading

Active

Can a faculty member disagree with a grading model precisely? Agreement is easy to measure, and the wrong target.

Explainability
Education
Human-AI interaction

Sim-to-real transfer

Ongoing

The reality gap lives in dynamics, observation, actuation and timing. Measure what can be measured; randomize only the rest.

Simulation
Domain randomization
System identification

Output

What has come out of it.

We publish the work and we teach it. Students in our research programme are co-authors where the work earns it, not acknowledgements.

Papers at top-tier venues

Collaborative work with our research cohorts, accepted at international AI conferences.

Physical AI demos deployed

Learned policies taken from simulation onto real hardware, end to end.

AIR 11

GATE CS

Rigorous mentorship and real deeptech exposure, in the same building as the research.

A full publication list with venues, abstracts and PDFs is going up here. Want a copy in the meantime?

Request the list

Research → product

The lab and the product
are the same argument.

We did not build a grading product and then look for a research story. The interpretability question came first; GunanQ is what our answer looks like when a real examination cell has to rely on it.

Assessment AI