Studying CS, doing competitive programming, and exploring machine learning and deep learning. Getting into agentic AI lately β building with LangChain, LangGraph, CrewAI, and learning how to put these pieces together into real systems.
- π Google Gen AI Hackathon Winner
- β‘ LeetCode β max rating 1802
- π Based in India
Languages
AI / ML
Backend & Tools
Vector Databases
Research project β given an audio clip of environmental sounds (waves, birds, traffic), find the matching aerial image, and vice versa. Adapted the SARCI framework for natural audio and improved Avg R@1 from 61.83% to 77.45% by introducing image masking during training and a joint loss function.
PyTorch SARCI CNN14 ResNet18 ADVANCE Dataset
Implemented Low-Rank Adaptation (LoRA) manually in PyTorch β no libraries, just the math. Injected trainable rank-decomposition matrices into Vision Transformer attention layers, reducing trainable parameters by 99.75% (86M β 221k) while successfully adapting an ImageNet model to satellite imagery classification.
PyTorch Vision Transformers LoRA EuroSAT
Production-grade support ticket routing system built from scratch. Reads customer tickets, searches 791 support documents using FAISS vector search, and decides whether to answer or escalate β with full adversarial robustness against prompt injections, jailbreaks, and encoded attacks.
Python FAISS LangChain Gemini RAG