At NexDoc.AI Inc., I designed reusable agentic pipelines for research, analysis, and content generation using LangGraph and Pydantic AI. I also implemented a standalone Visualization Agent and connected it to the chat interface and Asset Generation Pipeline.
I developed tools for a social media agent that aggregates cross-platform profile data for influencer discovery, and added map-grounded competitor analysis for local business research. I benchmarked LLM architectures and prompting strategies to inform internal model-selection guidelines.
For my E-Commerce Recommender System project, I built a retrieval-and-ranking pipeline using PyTorch, LightGBM, FAISS, and FastAPI. It reached 0.0293 recall@10, a 184% improvement over ALS, and I served the fixed pipeline with a parity test for online features.
I’ve also built projects in email campaign optimization, Ayurvedic wellness assistance, and multimodal document question answering. My hackathon work includes an entity-resolution pipeline, a stroke-risk model, and a playable 3D narrative game.
