About
Passionate AI/ML Engineer specializing in neural architecture design, transformer models, and production ML systems. Expert in developing machine learning models, building scalable data pipelines, and deploying AI solutions that learn and adapt.
Currently pursuing B.Tech in Data Science and AI at IIT Bhilai, where I serve as Coordinator of the DSAI Club.
My mission: reducing loss functions in both code and real-world problems.
Education
Indian Institute of Technology Bhilai
B.Tech. in Data Science and AI
Aug 2023 – May 2027 (Expected)
Relevant Coursework: Data Structures & Algorithms, ML, NLP, Deep Learning, DBMS, Statistics, OS
Published Open Source
pytorch-dml
Deep Mutual Learning Library
Overview
Production-ready PyTorch library for Deep Mutual Learning (DML) and collaborative neural network training. Implements state-of-the-art techniques where multiple neural networks train together and learn from each other's predictions, achieving 2-5% higher accuracy compared to independent training.
Quick Start
pip install pytorch-dml
from pydml import DMLTrainer
from torchvision import models
models = [models.resnet18(), models.resnet18()]
trainer = DMLTrainer(models, device='cuda')
trainer.fit(train_loader, val_loader, epochs=100)
Package Contents
Experience
Team Lead — Autonomous Financial Intelligence Platform
- Led an 8-member team to build a multi-agent financial intelligence system, generating 20% returns with 5–8% max drawdown over a 3–4 month backtested and paper-traded horizon.
AI Developer Intern
- Developed multi-agent automation and cloud APIs on AWS/GCP, cutting manual effort by 40%, boosting throughput by 30%, and maintaining 99.9% uptime while reducing infrastructural costs by 25%.
Projects
Kerala Ayurveda Medical RAG System
Stack: GPT-4 • LangGraph • CRAG • ChromaDB • FastAPI • Streamlit
Built a production-grade multi-agent medical RAG system using GPT-4, LangGraph, and CRAG with hybrid BM25+vector retrieval, reducing hallucinations to <10% (RAGAS) and improving retrieval accuracy by 35%.
Git Repository →Student Performance Enhancer
Stack: Ensemble ML • SHAP • Streamlit • MLflow
Engineered v2 of Kaggle Student Performance Enhancer with a modular ML pipeline; optimized ensemble models (R² 0.989), integrated SHAP explainability, deployed Streamlit app, and enabled MLflow tracking with versioned models.
Git Repository →PagedAttention Transformer
Innovation: Memory-Efficient KV Cache (60-80% reduction)
Implemented block-based attention mechanism with Copy-on-Write for beam search. Optimized inference pipeline achieving 2x throughput on large language models.
Git Repository →Academic Projects
Reciprocal Contextual Poetry Generation
Course: Machine Learning (CS 550)
Designed a modular experiment framework combining PPLM steering, a lightweight RLHF proxy (reward-model + supervised fine-tune), and hybrid inference-time control for generating context-aware poetry. Built reproducible pipelines for training, evaluation, and batch sweeps with aggregated metrics such as reward, perplexity, and distinctness.
View Repository →Bomberman AI — Intelligent Agent-Based Game (Tkinter)
Course: Artificial Intelligence (CSL304)
Built a full Bomberman game in Python with AI-driven bot agents capable of strategic movement, bomb placement, and real-time decision-making. Implemented A* pathfinding, state-based behavior (search, chase, evade), bomb safety simulation, and intelligent evasion scoring to create adaptive, challenging opponents.
View Repository →Skills
Languages
- Python
- SQL
- C++
ML / AI
- PyTorch
- TensorFlow
- Scikit-learn
- Hugging Face
- LangChain
- RAG
- Transformers
LLMs & Agentic AI
- GPT / BERT
- Fine-tuning
- Prompt Engineering
- Multi-Agent Systems
- Memory Management
MLOps & Explainability
- MLflow
- SHAP
- DVC
- Model Versioning
- Experiment Tracking
Tools & Cloud
- Git & Docker
- Linux
- GCP
- Streamlit & Jupyter
- Tableau
- FAISS & Vector DBs
- REST APIs & CI/CD
Positions of Responsibility
Coordinator — Data Science & AI Club
Led workshops, study groups, and hands-on ML sessions; coordinated events and mentoring to promote applied ML, Kaggle competitions, and research-oriented projects.
Team Lead — Hospital Location Predictor (Hackathon)
Led a team to build a geographic intelligence app that predicts optimal hospital placement by analyzing proximity to existing healthcare facilities, demographics, and regional infrastructure data.
Achievements
- Silver Medal — MITSUI & CO. Commodity Prediction Challenge (Kaggle): Ranked 36th out of 1,711 teams by building stable commodity return forecasting models using LME, JPX, US equity, and Forex market data.
- GQ Volatility Forecasting (Kaggle): Ranked 34th out of 386 participants forecasting 10-second-ahead Ethereum volatility using high-frequency order-book snapshots and cross-asset signals.
- Amazon ML Challenge 2025: All-India Rank 278 (score: 46.649) among thousands of participants.
- Extra ML Competitions: ranked 10 in the Humidity Prediction Challenge and 14 in the AI Meets Job Security Challenge, building end-to-end ML pipelines for intermediate-level forecasting and classification tasks.
- Hackathon — Hospital Location Predictor: Built a geographic intelligence app that predicts optimal hospital placement by analyzing proximity to existing healthcare facilities, demographics, and regional infrastructure data.
- Top 10% — Data Science Bootcamp (IIT Guwahati): Ranked among 500+ participants.
- Technical Writing & Knowledge Sharing: Authored 70+ ML/AI articles on Medium with 800+ monthly readers.