Built a pipeline that queries PubMed and extracts parameter-specific clinical data from research abstracts. Improved entity extraction F1-score to 90% by fine-tuning a PubMedBERT-based NER model on a manually annotated dataset. Implemented a distributed ML pipeline using Ray, PyTorch, and HuggingFace; deployed with Next.js and FastAPI.
- Rishikesh Kumar, Plant Disease Classification using State-of-the-Art Deep Learning Models, International Journal of Computing, Vol. 23, No. 4, 2024. DOI: 10.47839/ijc.23.4.3542.
Multi-agent cybersecurity news platform with two-round LLM deliberation and Borda-count ranking. Course project for IE 624 under Prof. Manjesh K. Hanawal.
NLP pipeline using double domain-adaptation and Integrated Gradients explainability. F1: 0.7778. Course project for CS 772 under Prof. Pushpak Bhattacharyya.
AUROC 99.57%, FPR 1.73% on ResNet50 / Tiny-ImageNet-200. Course project for IE 506.
Implemented PPO-max to improve RLHF training stability for LLMs; demonstrated gains over SFT baselines. Course project for IE 620.
Autumn 2024
Spring 2025
Autumn 2025
- IE 685: M.Tech Research Project