About
AI/ML Engineer with projects in fraud classification, energy forecasting, and LLM-powered applications. Built reproducible ML pipelines with LightGBM, XGBoost, and DVC, plus FastAPI scoring services and Qdrant-based RAG.
Projects3
FraudGuard
Developed a fraud classification system that achieved 0.540 Average Precision and 0.781 fraud recall on an 88,581-row chronological holdout set using LightGBM. Engineered 392 train-fitted frequency features and applied cost-aware thresholds. Built a React and FastAPI based fraud console for CSV/JSON scoring and sanitized prediction dashboards. Implemented SHA-256 verified model bundles, manifest checks, and readiness gating for release integrity.
- Python
- LightGBM
- Optuna
- MLflow
- DVC
- Docker
- FastAPI
- GitHub Actions
- React
WattPredictor - Energy Demand Forecasting
Created hourly electricity-demand forecasts across 11 NYISO zones using XGBoost and LightGBM. Engineered 672-hour lag features and performed model selection with 3-fold TimeSeriesSplit grid search. Automated hourly forecasting with a structured three-stage DVC pipeline and monitored model drift using Evidently AI. Packaged a non-root, multi-stage Docker image for deployment.
- Python
- XGBoost
- LightGBM
- DVC
- Streamlit
- Docker
- Evidently AI
- GitHub Actions
SmartRoute-AI
Built a user-scoped Qdrant retrieval augmented generation (RAG) system with FastEmbed, achieving 87.18% Recall@5 and 83.76% Precision@5 in local tests. Evaluated a LightGBM classifier with 96.67% accuracy and 96.66% macro F1 on 30 held-out questions. Packaged chat UI and API in a single Docker image with Server-Sent Events streaming, including provider retries, circuit breakers, Redis budget gates, and SSE error handling for robust operation.
- Python
- LightGBM
- LangChain
- React
- FastAPI
- Docker
- PostgreSQL
- Redis
- Langfuse
- Qdrant
Education1
B.Tech
2021 — 2025Dr. MGR Educational and Research Institute
Computer Science Engineering (Data Science and Artificial Intelligence)