Bangla NER classifier
State-of-the-art Bangla named entity recogniser serving 10 entity classes at 90.49% macro F1, deployed with FastAPI and TensorRT for the national Bangla corpus platform.
- PyTorch
- XLM-RoBERTa
- FastAPI
- TensorRT
- Node.js
Senior Associate Engineer – AI at ZS Associates
I design agentic AI systems for pharma R&D and lead the teams that take them to production.
I am an AI engineer with almost five years of experience taking models from research to production across LLM agents, NLP and computer vision. At ZS Associates I design agentic AI for pharma R&D and lead development of a production suite of agentic document tools for a global pharmaceutical sponsor. I also research low-resource Bangla NLP, with papers at NAACL 2025 and EMNLP 2023.
State-of-the-art Bangla named entity recogniser serving 10 entity classes at 90.49% macro F1, deployed with FastAPI and TensorRT for the national Bangla corpus platform.
Extractive QA for Bangla. Given a passage and a question it returns the answer span, reaching 83.71% F1 on SQuAD-bn, 6 points above the previous best.
Demo app from the BanglaBait research. Paste a Bangla headline and it tells you whether it is clickbait, with a probability score, using a transformer trained in a GAN framework.
How Sparse Transformers cut the quadratic cost of attention by factorising it into strided and fixed patterns, and what that buys you in context length.