AI Engineer at Meltwater. I build production-grade agentic pipelines and NLP systems. Previously a Data Scientist at American Express · IIT Madras (B.Tech + M.Tech, Data Science)
- Agentic pipelines: multi-agent data creation, from scope and labeling rules through a versioned spec
- NLP systems: multilingual models, fine-tuning, and calibration for production classification
- Evals: held-out benchmarks, LLM-as-judge, and autoresearch loops
- Retrieval and serving: RAG, reranking, and ONNX models on AWS
- Built an agentic data-creation pipeline that freezes a versioned labeling spec for a platform and a profile field
- Author-country model on a 251-way task: 94.7% accuracy and ECE 0.008 on 31,153 held-out authors
- GenAI Lens skip model at 94% F1, served with ONNX on Lambda
- Three consecutive Meltwater hackathon wins · AmEx Modeling Super Bowl top 5 of 272 · Analytics India 1st of 225
| Project | What it is | Stack |
|---|---|---|
| Regulation RAG Assistant | Cited Q&A over EU banking regulation with ColBERT and HyDE | ColBERT, HyDE, Weaviate, Streamlit |
| Startup ecosystem RAG | Document chat with proposition-level retrieval and an agentic rerank | Ollama, Chroma, Streamlit |
| GuessWars | A transformer that plays Hangman, trained with self-play | PyTorch |
Python PyTorch Hugging Face LoRA Transformers RAG ColBERT LangChain FastAPI Streamlit ONNX XGBoost Claude AWS Docker
- pytorch-forecasting #2442: add a usage example for xLSTMTime
- pytorch-forecasting #2445: add a DLinear v2 tutorial notebook