At the Indian Institute of Science Education and Research (IISER), Bhopal, I built a PyTorch federated LLM fine-tuning pipeline across 50 non-IID clients, reducing training loss by 79% over 100 rounds. I also developed an OLS-based semantic alignment method to mitigate stale-update drift in asynchronous federated learning.
I applied LoRA to reduce communication payloads by about 1000× while enabling efficient LLM fine-tuning. In my FinSight project, I analyzed customer records and developed a risk-aware offer decision framework, validating the targeting approach with A/B testing.
I also developed a computer vision narration system that combines OCR and text-to-speech for accessible scene narration. It achieved 90.96% OCR accuracy and was published at IEEE IC-CGU 2025.
