AI & ML undergraduate at BIT Mesra · Building AI systems from model to deployment
I work on agentic applications, LLM fine-tuning workflows, and machine learning research. I enjoy making models useful through reliable APIs, evaluation, and observability.
- 🎓 B.Tech in Artificial Intelligence and Machine Learning at Birla Institute of Technology, Mesra (2023–2027).
- 🔬 Research experience in camera-based heart-rate estimation with physics-informed learning.
- 🛠️ Interested in agent reliability, LLM evaluation, model serving, MLOps, and practical full-stack AI products.
- 🤝 AI & Quadcopter Head at the Aerospace Society and Outreach Head at the Society of Data Science, BIT Mesra.
LangGraph · FastAPI · React · PostgreSQL · Redis · Gemini · Opik · Google Cloud Run
An interactive customer-support demo for order questions and refund requests. A tool-using agent gathers customer-scoped data, while deterministic application code enforces refund policy, ownership, and ledger writes. The app includes a live audit timeline, admin console, model fallback, evaluation cases, and Opik traces.
Live app · Project overview · Code
PyTorch · Hugging Face Transformers · QLoRA · MLRun · AWS SageMaker · vLLM · Terraform
A portfolio implementation for analyzing non-disclosure agreements with an open-weight LLM. It prepares ContractNLI data, fine-tunes Hermes-4-14B adapters with QLoRA, evaluates hypothesis labels and supporting clauses, and includes model registry, serving, monitoring, and infrastructure code. The repository documents recorded experiments and the setup needed to run its AWS/MLRun workflows; its full production path is still being completed.
Code and architecture · Training and evaluation
Python · PyTorch · OpenCV · MediaPipe · FastAPI · React · Docker
A live camera-based rPPG application for contactless heart-rate estimation. PRISM extracts video quality and physiological features, predicts which specialist model should perform best for the input, and routes inference to that model. The React interface and FastAPI backend bring the research pipeline into an interactive demo.
As part of my BIT Mesra research, I developed the physics-informed model selection approach and reported 1.08 BPM MAE in motion-intensive evaluation. The work contributed to a patent filing and a paper submitted to IEEE TCBB.
Live app · Code and research overview
| Area | Tools and concepts |
|---|---|
| Languages | Python, C++, SQL |
| Agents and generative AI | LangGraph, LangChain, tool calling, ReAct agents, RAG, Hugging Face Transformers |
| LLM training and evaluation | PEFT/LoRA, QLoRA, PyTorch DDP, vLLM, DeepEval, Opik, Promptfoo |
| ML and data | PyTorch, scikit-learn, pandas, NumPy, feature engineering, model evaluation, Kafka, Spark |
| Backend and data stores | FastAPI, REST APIs, PostgreSQL, Redis |
| Cloud and MLOps | AWS, Google Cloud, Docker, GitHub Actions, MLRun, Terraform, Kubernetes |
Research Intern · BIT Mesra · Jan–Apr 2026
Developed a physics-informed meta-learning approach for rPPG heart-rate estimation. Reported 1.08 BPM MAE in motion-intensive evaluation; contributed to a patent filing and a paper submitted to IEEE TCBB.
Freelance Machine Learning Engineer · Independent client · May–Jun 2026
Built a predictive-maintenance pipeline and FastAPI inference service with Kafka, Spark, and monitoring. Reported 0.94 F1 for failure-risk prediction.
Flipkart GRID 8.0 semifinalist (2026) · MMVY scholarship recipient (three years) · G.P. Birla Scholarship recipient