12th standard

Bachelor of Engineering - BE

Bachelor's degree

Selected for the ELC Summer intensive research program under the Computer Science and Engineering Department at Thapar Institute.

1. Designed a multimodal CNN and DNN fusion pipeline combining MPU9250 IMU tremor signals via Arduino serial with Librosa voice features

2. Signal preprocessing modules raised combined model accuracy from 88% to 91% on clinical Parkinson's Disease data

3. Benchmarked unimodal vs fused architectures across 5 feature configurations

4. Identified MFCC-delta augmentation as the single largest accuracy driver, contributing +1.8% over the raw MFCC baseline

ML Research Intern, Parkinson's Early Detection

Designing a physics-guided deep learning framework for accelerated MRI reconstruction under the supervision of Dr. Anurag Tiwari at Thapar Institute.

1. Designed an 8-stage physics-guided cascade network with a BCE-supervised MLP gating scalar that softly blends dedicated knee and brain U-Net branches per sample

2. Enforces Fourier-domain data consistency at every stage without requiring anatomy labels at inference

3. Achieved peak SSIM 0.757 on mixed fastMRI 4x acceleration, outperforming single-branch baselines by +1.78% SSIM at 302 GFLOPs and 112ms per slice

4. Competitive with SwinMR and HUMUS-Net trained on single anatomies

Status: Paper authored

Lead Author, Dual-Branch MRI Reconstruction Research

Team research project on multimodal fusion architecture for autonomous search and rescue in visually degraded environments, under Dr. Ankit Soni at Thapar Institute. Collaborating with @Navya Abrol.

1. Surveyed 50 sensor fusion and SLAM systems across thermal, acoustic, LiDAR, radar, and visual modalities

2. Covered the full SLAM pipeline behaviour in degraded SAR environments

3. Identified 8 cross-cutting failure patterns and proposed degradation-aware fusion architectures

4. Building follow-up multimodal fusion system using Isaac Sim and ROS 2

Status: Review paper authored. Follow-up implementation in progress. Target venue: IROS and ICRA

Co-Author, Multimodal SAR SLAM Survey

Applied AI & Robotics Engineer. Developing ROS 2 autonomous systems, autonomous LLM agents, and PyTorch vision pipelines

Know about Kunal Somani's work experience, education, bootcamps, certifications, and much more. Kunal Somani