I've built AI systems spanning meeting intelligence, retrieval-augmented question answering, and GPU-accelerated deep learning.
For Meeting AI Agent, I developed an assistant that turns meeting audio into diarized transcripts, structured summaries, and assigned action items. I built its asynchronous backend with FastAPI, Celery, PostgreSQL, Redis, Docker, and Google Calendar integration, alongside MLOps evaluation, monitoring, feedback, promotion, and retraining workflows.
My ICMR'25 publication proposed a RAG system for open-ended lifelog question answering using embedding retrieval and LLMs without fine-tuning, achieving 94.35% Recall@20 and 39.54% ROUGE-L on 14,187 QA pairs. I also implemented a CUDA autoencoder with custom convolution, pooling, and backpropagation kernels, increasing throughput 3.25× and reducing training time by 69%.
I've also led end-to-end development of a florist e-commerce platform with Next.js, MongoDB Atlas, REST APIs, and Cloudinary. I'm a five-time Academic Excellence Scholarship recipient and a Kaggle AI Mathematical Olympiad Silver Medalist.
