Research Projects

We propose a Graph Neural Network for detecting AI generated videos. By enforcing Neighborhood Consistency Regularization exclusively across temporal edges, our model prevents spatial overfitting to known artifacts. This structural constraint significantly improves out of distribution generalization, achieving state-of-the-art robustness on unseen generative models.

An adaptive VR badminton coaching system featuring high-fidelity aerodynamics, multi-stroke kinematics with Holt-linear state filtering, dynamic body-scaled target calibration, and full-body IMU tracking.

A bio-potential EOG device that enables motor-disabled users to control computers using eye movements, offering hands-free access to navigate and scroll digital interfaces.

An asymmetric asynchronous Mixed Reality teleconsultation platform bridging pre-surgical patient-clinician communication via intentional semantic abstraction, dynamic local-coordinate serialization, and interactive 6DOF organ models.

Immersive augmented reality navigation and 3D architectural exploration for the Tirumala temple complex.

Developing a model to do real time conversion of Indian Sign Language(ISL) to sentence.

Interactive virtual reality platform for teaching projectile motion and gravity concepts through immersive simulations.

Computer vision algorithm for transformation of Night time Images to corresponding Day time.

A modular standalone extended reality platform for upper limb stroke rehabilitation, featuring clinician-configurable tolerance corridors, dual-loop adaptation, and controller-free optical hand tracking on Meta Quest.

Comprehensive VR simulation platform for Crystal Structure.

Real-time computer vision pose estimation framework providing biomechanical joint-angle correction and feedback for unassisted yoga practice.

Interactive 6DOF virtual reality environment allowing students to visualize invisible 3D electromagnetic vector fields, manipulate charges, and assemble virtual circuits.

Deep learning embedded vision system running on edge hardware to recognize road sign typography under occlusions and provide low-latency spoken alerts to drivers.

Immersive 3D geometry sandbox enabling students to manipulate geometric solids, inspect intersecting planes in spatial coordinates, and learn spatial proofs interactively.

Semantic parser converting freeform clinical and exercise text instructions into synchronized 3D avatar animations for physical therapy.

A memory-efficient multi-tier retrieval framework indexing 21 million Wikipedia passages inside 2.58 GB active VRAM, allowing full co-location of an 8B LLM reader on a single consumer GPU.

When users wear always-on egocentric cameras (such as Meta Ray-Ban glasses), they accumulate hours of video daily. EgoScore operationalizes moment importance across social significance, emotional salience, novelty, and memorability by fusing six perceptual signals with Ebbinghaus-inspired temporal decay.

Investigating how algorithmic monoculture in automated hiring pipelines induces structural wage compression and correlated candidate rejection through agent-based macroeconomic simulations and two-stage cross-encoder re-ranking audits.

Dense retrieval systems encode documents into a single semantic vector, coupling factual content with logical implication. GraphCLaRa introduces a causal partition mask over distilled reasoning LLMs, enforcing architecturally disjoint reasoning and semantic sub-representations within a single forward pass alongside differentiable Personalized PageRank graph routing.

An autonomous, agentic research assistant and pedagogical paper coach built on decoupled React/FastAPI architectures, LangGraph, and MCP with Reflexion-style self-critique loops, literature cross-verification, and sandboxed interactive code execution.

A biologically inspired modular neural control architecture combining evolutionary phylogenetic hyperparameter optimization with lifelong ontogenetic local learning without global backpropagation through time.

Material-aware physical-artistic stylization engine for Meta Quest 3, bridging continuum mechanics with non-linear anime exaggeration and view-invariant non-photorealistic inking.