MSc Data Science @ ETH Zurich · Computer Vision & 3D Reconstruction · Deep Learning · MLOps
I'm a first-year MSc Data Science student at ETH Zurich, currently doing research at the Computer Vision and Geometry Group (CVG) on 3D reconstruction for in-the-wild indoor videos — working with transformer-based models like DUSt3R, MASt3R, and DAV3 to improve robustness where classical SfM methods fail.
Before ETH, I completed my B.E. in Computer Science at Coimbatore Institute of Technology (top 2% of class) and a Diploma in Data Science at IIT Madras — both in parallel. I've interned at Caterpillar Inc. (monocular depth estimation for autonomous trucks), Akamai Technologies (DevOps/ETL at scale), and Samsung R&D (generative models).
I genuinely enjoy hard ML problems — whether that's getting a model to run on edge hardware, figuring out why a vision transformer fails on certain inputs, or building systems that hold up at scale.
- 🔬 ETH CVG Research — Benchmarking ViT-based 3D reconstruction models (DUSt3R, MASt3R, DAV3); building hybrid SfM pipelines with COLMAP; multi-GPU training on Euler HPC
- 🧠 VLM Mechanistic Interpretability — Analyzing attention mechanisms in Qwen 3-4B across synthetic datasets to understand relational reasoning limitations (GitHub)
Languages: Python · C++ · Java · C · SQL · Shell Script
ML & Frameworks: PyTorch · TensorFlow · ONNX · Scikit-learn · NumPy · Pandas · Flask · Django
Focus Areas: Computer Vision · 3D Reconstruction · Transformers · Edge AI · NLP · RAG · Multimodal (Vision-Language) · Distributed Systems
Tools: Docker · Git · Jenkins · Azure DevOps · Kubernetes · AWS S3 · GCP · ClickHouse · PostgreSQL · MongoDB · Grafana · Linux
| Project | Description | Stack |
|---|---|---|
| Federated IDS — DBN | Privacy-preserving WiFi intrusion detection with federated learning — 98% accuracy on 300k+ samples. Springer 2025 🏆 | PyTorch · Go · gRPC · FedAvg |
| VLM Relational Reasoning | Mechanistic interpretability of Qwen 3-4B across 4 synthetic datasets — ablation, activation patching, circuit analysis | PyTorch · CLEVR · ChartQA |
| Doc-Chat RAG | End-to-end RAG pipeline for PDF Q&A, summarization, and quiz generation | Flask · ChromaDB · LLM |
| Music Streaming App | Full-stack streaming app with auth, playlists, and search | Flask · SQLAlchemy |
- 📄 Springer 2025 — Accepted paper: FEDDBN-IDS: Federated Deep Belief Network for Intrusion Detection
- 🏆 LeetCode Knight — Top 2.5% globally (Rating 2020+)
- 🥇 Winner — Internal Smart India Hackathon 2023 (1/16 teams)
- 🥈 2nd place — Innovision24 Hackathon (2/48 teams)
- 🔢 A. John Wilson Endowment Prize — Highest mark in Mathematics across departments
- 🌐 Open-source contributor to TheAlgorithms/Python (180k+ ⭐)