I am currently pursuing Masters in Computational Science and Engineering at Georgia Institute of Technology with a strong interest in Generative AI, 3D vision, and scalable AI systems.

I am working with Prof. Frank Dellaert as a contributor to the open-source Structure-from-Motion framework GTSfM. I integrated 3D Gaussian Splatting optimization into the pipeline for dense reconstruction leveraging Dask for distributed parallel computing. More recently, I have been exploring 3D foundational models to generate 3D point clouds and Gaussians to enable large-scale reconstruction with novel clustering and merging algorithms.

I am also a Graduate Research Assistant at the SocWeb Lab working with Prof. Munmun De Choudhury at the intersection of ML and mental health, evaluating modern ML methods, especially LLMs and their effect on mental health. My work has led to two papers:

• Lived Experience Not Found: Proposed a new methodology and benchmark to assess LLM performance in clinical decision-making for psychiatric treatment and harm reduction. The work was accepted as a Main Conference paper at NAACL 2025.
• Reasoning is Not All You Need: Created a diverse dataset with clinically grounded synthetic multi-turn mental-health conversations with the conversation flow inspired by prior psychological literature. The work was accepted as a Main Conference paper at ACL 2026.

Some previous projects and works:

• Robust Saliency Prediction: Worked with Prof. Matthias Bethge at the Tübingen AI Center and developed a robust saliency prediction model by incoporating Vision-Language Models (VLMs) surpassing state-of-the-art on fixation prediction benchmarks and improved generalizability. The work was accepted as a Highlight paper at ICCV 2025.

• Retrieval-Augmented Generation (RAG) based Agentic AI system: Built an end-to-end RAG based Agentic AI chatbot leveraging 7 years of proprietary conversation in the ethanol trading space and productionalized the application by deploying it on Azure using Dockers and Kubernetes to ensure modularity, security and scalability.

• Unlearning in LLMs: Fine-tuned Alpaca 7B model to reduce bias in its responses using reinforcement learning inspired DPO loss variant.

• DCC: Trained a Computer Aided Diagnosis system to segment colon cancer from Whole Slide Images using CNN-Vision transfomer hybrid.

• NeuroNet: Developed a custom model archictecure to classify patients with neurodegenerative from videos using a feature fusion from 3D CNNs and handcrafted features.

• 🔭 I’m currently working on GTSfM to generate 3D reconstruction for large datasets.

• 👯 I’m looking to collaborate on stuff in the Perception, 3D vision and building ML systems domain.

• 📫 How to reach me: Please reach out to me either on email or linkedin profile.

• 😄 Pronouns: he/him