> Terminal Active
M.Tech Data Science & AI candidate at IIT Madras. Journeying from the heights of Ladakh to building scalable machine learning pipelines and production-ready deep learning systems.
I am an M.Tech student in Data Science and Artificial Intelligence at IIT Madras, driven by a passion for turning raw data into actionable intelligence. My journey ranges from high-altitude schooling in Kargil to completing my B.Tech in CSE at NIT Silchar.
Today, my work focuses on Machine Learning, Deep Learning, and Generative AI—specifically LLMs, RAG, and Agentic AI systems. I specialize in building end-to-end Python workflows, from data preprocessing and feature engineering to deployment-oriented pipelines. Ultimately, my goal is to bridge AI research and ML engineering, ensuring models don't just work in Jupyter notebooks, but thrive in production.
Tech Stack
Education
A fully reproducible biometric identification pipeline using DVC. Built a 4-stage automated workflow (Prepare → Transform → Train → Evaluate) achieving 0.92 Macro F1 using MobileNetV2 with TTA.
Developed an agentic Retrieval-Augmented Generation (RAG) system for multi-turn, document-grounded question answering. Orchestrated intelligent agents using LangGraph and integrated FAISS vector search.
View GitHubExplored factor-driven success in TMDB metadata. Managed data cleaning by treating 0-values as NaN and used Log-transformations to uncover hidden trends in right-skewed revenue data.
Implemented a Logistic Regression classifier to predict heart disease likelihood. Leveraged health markers like cholesterol and BP to enable early medical intervention.
Configurable MLP engine built using pure NumPy. Features backpropagation, Xavier init, and 6 optimizers (Adam, Nadam, etc.). Tracked experiments via Weights & Biases.
Deep ResNet-MLP pipeline for prompt-response fitness scoring. Features KL Histogram Loss for matching score distributions and SWA/EMA for training stability.
Ensemble regression suite benchmarking Bagging, Boosting, and Stacking. Captures temporal patterns via Sine/Cosine cyclic feature encoding.