I'm a graduate student at Northeastern University pursuing my MS in Electrical and Computer Engineering, specializing in Computer Vision and Machine Learning. Previously, I worked as an AI Engineer at Checkit Analytics, where I designed and deployed LLM-based RAG pipelines and fine-tuned open-source models for production financial analytics — reducing external API dependency and shipping AI products end-to-end.
I am doing research under the supervision of Prof. Sarah Ostadabbas in the Augmented Cognition Laboratory (ACLab), focused on video understanding. ACLab builds intelligent systems that can reason about the real world from limited data rather than from ever-larger datasets. The lab specializes in motion-centric video understanding — because motion captures causality, intent, and dynamics that static frames cannot — and develops data-efficient models for tracking, behavior analysis, and action recognition in small-data domains where labels are scarce, expensive, or hard to share. My work sits in that agenda: learning structured representations such as pose, trajectories, and temporal dynamics so vision systems can interpret human and animal behavior in healthcare, robotics, and other real-world settings.
My background spans Machine Learning, Computer Vision, and Large Language Models (LLMs). I've published peer-reviewed research on COVID-19 detection from chest X-rays, including model compression for edge deployment, and built production AI systems spanning healthcare, financial analytics, and misinformation detection. These experiences shaped how I think about building AI that is not just accurate, but practical, reliable, and deployable.
I'm particularly interested in problems that require owning the full ML lifecycle — from data and experimentation to evaluation, deployment, and monitoring at scale.
in arXiv preprint arXiv:2609.09528, 2026.
in ICBSII 2024, Chennai, India.
[PDF]
in Advances in Computational Intelligence and its Applications 2023, Taylor & Francis / CRC Press.
[PDF]
Checkit Analytics · California, United States
Northeastern University · Boston, MA
Centre of Advanced Research in Electrified Transportation · Aligarh, India
Indian Institute of Information Technology · Allahabad, India
Paper (arXiv) · Code (GitHub) · Dataset (Hugging Face)
Contrastive video benchmark isolating speed, magnitude, and direction — 82 self-recorded clips and 240 Q&A items, evaluated across 170+ configurations spanning 8 video-language models, 4 frame selectors, and 5 frame budgets. Open models score at the 6.25% chance floor on motion despite reaching 58–70% on standard video QA.
Multi-agent agentic AI system automating job search, outreach, tracking, and interview prep. LLM agents for resume tailoring and personalized recruiter messaging with stateful human-in-the-loop controls.
End-to-end healthcare AI for medical report summarization, diagnostic image analysis, and risk prediction. RAG Q&A pipeline with MLOps workflows, monitoring, and bias checks.
Facial emotion recognition across 7 expressions on FER2013, using a custom CNN and ResNet-18. A CLAHE contrast-enhancement pipeline improves robustness on low-contrast images, evaluated baseline vs. enhanced preprocessing — 58.19% test accuracy with the custom CNN.
Custom CNN achieving 97.2% accuracy in COVID-19 classification from chest X-rays. Model compression for edge deployment with clinical-level detection metrics.
What I learned deploying retrieval-augmented generation at scale — chunking strategies, embedding models, and the hard trade-offs nobody talks about.
How we replaced expensive API calls with a fine-tuned open-source model — data prep, training loops, and deployment on AWS.
An end-to-end walkthrough of implementing SLAM on a mobile robot — from sensor fusion with Kalman filters to real-time map building.
How I designed a custom CNN achieving 97.2% accuracy on CXR classification, compressed it for edge deployment, and presented findings at two international conferences.
Open to ML / computer vision roles, research collaborations, and robotics projects.