Curriculum Vitae

IIT Kharagpur · Kharagpur, India

Causal Reinforcement Learning / World Models

• Research under Prof. Dibbendu Roy on causal reinforcement learning and world models, building on prior work in causally-aware policy optimization
• Developing a framework to evaluate the causal capabilities of world models across Pearl's Ladder of Causation — association, intervention, and counterfactual reasoning
• Designing a three-stage evaluation pipeline using CausalGym and custom environments to quantify causal understanding and intervention awareness in learned latent representations
• Ongoing research submitted towards publication at AAAI 2027

University of Groningen & IIT Indore · Groningen, Netherlands / Indore, India

Speech Synthesis / Evaluation

• Joint research under Prof. Shekhar Nayak (University of Groningen) and Prof. Nagendra Kumar (IIT Indore) on the reliability of objective evaluation metrics for in-the-wild text-to-speech systems
• Studied whether existing objective MOS predictors can reliably evaluate speech synthesized by discrete-token and in-the-wild TTS models
• Designed pipelines for speech generation, human listening studies, statistical significance testing, and metric benchmarking
• First-authored paper accepted at INTERSPEECH 2026 (main track)

IIT Indore · Indore, India

Reinforcement Learning / Causal Inference

• Developed causal variants of REINFORCE, PPO, and A2C using causal-aware, action-independent baselines for variance reduction in policy gradient estimation
• Derived theoretical regret guarantees and validated the methods on custom Structural Causal Model (SCM) environments
• Demonstrated improved sample efficiency and policy learning stability under causal interventions and distribution shift
• Extended first-authored paper based on this work submitted to AAAI 2027

IIT Indore · Indore, India

Machine Learning / Communication Systems

• Teaching assistant for EE216: Machine Learning and Signal Processing and EE319: Communication Systems
• Designed assignments, quizzes, and tutorial material on deep learning, signal processing, probability theory, Markov chains, and queueing systems
• Ran tutorials and problem-solving sessions
• Assisted with evaluation and grading

Samsung R&D Institute India · Bengaluru, India

6G Standards / Deep Learning

• Built deep learning frameworks for the 6G Standards Team, focused on high-fidelity CSI-RS (Channel State Information Reference Signal) recovery
• Designed UNet-inspired models for spectral super-resolution, reaching a normalized mean squared error of 1e-3 on sparse, complex-valued 2D-FFT signals
• Co-authored an innovation currently under internal patent filing
• Secured an Advanced Developer return offer

OPTIMAL Research Group, IIT Indore · Indore, India

Audio-Visual Speech Enhancement

• Developed LSTMSE-Net, an audio-visual speech enhancement model that isolates and enhances speaker audio in noisy environments
• Engineered a temporal feature extraction pipeline using RNN and LSTM units to jointly model audio-visual dependencies
• Achieved a 3x reduction in inference time over the baseline alongside improvements in speech quality
• Paper accepted at the COG-MHEAR AVSEC workshop, INTERSPEECH 2024

LIPG, IIT Indore · Indore, India

Medical Image Segmentation

• Developed U-Net based models for liver tumor segmentation, handling pre-processing and custom callbacks, metrics, and loss functions
• Implemented Squeeze-and-Excitation networks and Atrous Spatial Pyramid Pooling, reaching 98% segmentation accuracy
• Co-authored papers published in Elsevier Biomedical Signal Processing and Control

Northwestern University · Illinois, USA

Medical Image Segmentation

• Implemented liver tumor segmentation models including DeepLabv3+, UNet, and HiFormer-L on the LiTS dataset in PyTorch and TensorFlow
• Engineered modular PyTorch data loaders and transformation pipelines for LiTS
• Source code released at github.com/TheAlphaJas/Liver-Tumor-Seg-Implements