AI/ML Engineer | Kggle Expert | Building production systems that scale | Training models at 3 AM

Currently pursuing B.Tech in Data Science & AI at IIT Bhilai

I build AI systems that work in production, not just in notebooks. My focus is on LLMs, multi-agent architectures, and making ML models actually deployable at scale.

• Currently Building: Human Slop — Anti-AI social platform with hardware-bound authentication
• Published: pytorch-dml — Deep Mutual Learning library on PyPI
• Writing: 70+ ML/AI articles on Medium with 800+ monthly readers
• Recent Won: Silver Medal on Kaggle (Rank 36/1,711) in MITSUI Commodity Prediction Challenge

Production-ready PyTorch library for Deep Mutual Learning and collaborative neural network training.

• Impact: 2–5% accuracy improvement over independent training
• Scale: 7,000+ lines of code, 34 modular components
• Features: AMP, DDP, ONNX export, full test coverage
• Stack: PyTorch, CUDA, Distributed Training

Privacy-first social platform that blocks 100% of AI-generated content using typing forensics.

• Architecture: Dual-database system, hardware-bound biometric auth
• Innovation: Real-time keystroke analysis (WPM, burst patterns, pauses) → Human Score
• Stack: FastAPI, React Native, Supabase, Device Fingerprinting
• Status: Fully deployed (Web + Mobile)

Multi-agent AI system for autonomous market analysis and portfolio management.

• Performance: ~20% returns with 5–8% max drawdown (backtested + paper traded)
• Architecture: 7-layer system with temporal data fabric, RL forecasting, agentic debate framework
• Team: Led 8-member engineering team
• Stack: Pathway, PyTorch, LangChain, Real-time streaming

Production-grade multi-agent medical information retrieval system.

• Accuracy: <10% hallucination rate (RAGAS), 35% retrieval improvement
• Architecture: CRAG framework with hybrid BM25 + vector retrieval
• Stack: GPT-4, LangGraph, ChromaDB, FastAPI, Streamlit

Contributions merged upstream after maintainer review. Focused on delivery integrity, LeRobot import paths, and bucket-capable catalog/import flows.

Stack: Python, MCAP, DuckDB/Parquet, CLI, storage roots (local + object store)

Team Lead — Autonomous Financial Intelligence Platform
Independent Research Project, IIT Bhilai | Oct 2025 – Dec 2025
Led 8-member team building multi-agent financial AI system combining real-time streaming, RL, and LLM reasoning.

AI Developer Intern — Kartavya Technology
Remote | Jun 2024 – Aug 2024
Built multi-agent automation systems and cloud APIs on AWS/GCP. Reduced manual effort by 40%, maintained 99.9% uptime, cut infrastructure costs by 25%.

Coordinator — Data Science & AI Club, IIT Bhilai
Led workshop,and hackathons promoting ML culture within the college.

Languages: Python, C++, SQL, JavaScript, Bash

ML/AI: PyTorch, TensorFlow, Scikit-learn, Hugging Face, XGBoost, OpenCV

LLMs & Agentic AI: GPT, LangChain, LangGraph, RAG, Fine-tuning, Prompt Engineering, Multi-Agent Systems

MLOps: MLflow, DVC, SHAP, Model Versioning, Experiment Tracking, Docker, Kubernetes

Infrastructure: GCP, AWS, CUDA, FastAPI, REST APIs, CI/CD, Distributed Training

Data: PostgreSQL, MongoDB, Neo4j, FAISS, ChromaDB, Pinecone

• Kaggle Silver Medal — MITSUI & CO. Commodity Prediction Challenge (Rank 36/1,711)
• Kaggle Top 10% — GQ Volatility Forecasting (Rank 34/386)
• Amazon ML Challenge 2025 — All-India Rank 278
• Winner — Pixel Perfect Hackathon (IIT Bhilai) — Improved baseline by 23%
• Top 10% — Data Science Bootcamp (IIT Guwahati) — Ranked among 500+ participants
• Technical Writer — 70+ articles on Medium, 800+ monthly readers
• YC Startup School India — Bengaluru 2026

"Reducing loss functions in both code and real-world problems."

If you find my work valuable, consider starring my repositories!