I'm an AI/ML Researcher and Production Engineer building reliable machine learning systems and studying how to make AI safer, more capable, and useful in the real world.
I didn't always believe AI could make a real impact. Then I saw it save me hours of work, and I realized this technology is going to be bigger than anyone imagines. That curiosity turned into a mission: build AI systems that are not just impressive in notebooks, but reliable, safe, and genuinely useful in production.
Today, I work at the intersection of AI research, production ML, and agentic systems:
- Starting M.S. by Research in Data Science & AI at IIT Madras - advised by Dr. Krishna Pilutla and Prof. Balaraman Ravindran, under the India AI Mission
- Technical Operations Engineer at Secure AI Futures Lab / SteadRise - AI safety research, data systems, and graph engine development
- Top-Rated AI/ML Developer on Upwork - delivering end-to-end RAG chatbots, forecasting, and agentic automation for global clients
- Developer at ML Hub - bug-fixing before launch and implementing AI features
- Shipping AI products on Upwork - RAG chatbots, semantic search, and automation workflows
- Exploring agentic AI with LangGraph, CrewAI, and evaluation-driven development
- Studying privacy attacks and defenses in federated learning
- Building production ML forecasting systems for renewable energy at scale
- Researching AI safety for multilingual and low-resource contexts
|
B.Tech Thesis - Privacy vs. Performance Empirical study of machine unlearning in recommendation systems. Compares SISA retraining with Gradient Ascent across Matrix Factorization and SASRec on MovieLens-1M. Evaluates with MIA AUC, Jaccard@10, and recommendation drift.
|
BlockVerse - HR AI Product Personalized AI mock-interview platform with role-specific scenarios, generated scripts, voice-over, AI visuals, and video simulations. Built with OpenAI, Streamlit, and Docker.
|
|
BlockVerse - HR Analytics Interview-feedback sentiment analyzer using Whisper transcription and GPT-3.5 analysis. Provides 1-5 sentiment scores, evidence quotes, detailed feedback, and downloadable reports.
|
Next.js + TypeScript + Framer Motion Personal AI/ML portfolio showcasing projects, experience, skills, and certifications. Built with Next.js, TypeScript, Tailwind CSS, and Framer Motion animations. Deployed on Vercel.
|
Click to see more projects
Production ML - NewGrid Consulting LLC
Operational 7-day power forecasting platform for ~450 solar and wind plants. 1,700+ deployed models with 98% deployment success. LightGBM/CatBoost/XGBoost ensembles, FastAPI inference, and automated weekly cron workflows.
SteadRise / Secure AI Futures Lab
Interactive graph exploration for 10k+ nodes with 5k connections. Provenance-aware ingestion pipeline handling 5M+ records. Layered views, path-finding, and LLM-as-judge RFP scoring.
Open-source - Omdena (10+ contributors)
Context-aware mental-health platform with RAG, Llama Guard safety guardrails, DPO optimization, and DistilBERT query classification. Deployed on Hugging Face.
Agentic AI - Personal Project
Intelligent web search agent with LLM query validation, ChromaDB semantic caching (50% similarity threshold), Playwright scraping, and Groq (Llama 3.1) summarization.
Upwork Client - Florida-based HVAC Company
End-to-end RAG chatbot integrated into company website. Automated business processes using Zapier, LangGraph, and MindsDB MCP server.
Upwork Client - Insurance
Premium prediction API using AWS, Pydantic, and FastAPI for an insurance client.
Jadavpur University ML Hackathon 2024
Few-shot learning, chain of thought, and advanced prompting to assess and improve LLM (Llama-2) performance in logical, analytical, and mathematical tasks.
AI Consultant - Jaagruk Bharat
Built a schema-matching system to find relevant government schemes from 50,000+ entries. POC for matching user profiles to scheme eligibility criteria using semantic similarity.
CI/CD, Model Registry, and Deployment Pipelines
Collection of MLOps implementations including Docker containerization, Kubernetes orchestration, MLflow tracking, Dagshub versioning, and Jenkins CI/CD pipelines.
- Multi-agent orchestration with LangGraph and CrewAI for complex research workflows
- AI safety evaluations for low-resource and multilingual contexts
- Production RAG systems with semantic caching, chunking optimization, and guardrails
- Federated learning privacy attack and defense mechanisms
- GNN pretraining / fine-tuning for unlearning experiments (ongoing research)
"I build AI systems that are reliable, safe, and actually useful in the real world."