Adarsh Bajpai

I’m an AI Engineer building real-world machine learning systems and pipelines from development to deployment, with a focus on performance, reliability, and practical impact. I thrive on taking models beyond experimentation—turning them into dependable services and workflows that users can actually rely on.

In my current role, I developed a RAG pipeline for document querying and question answering with support for Hindi text, using chunking strategies to preserve context and improve response quality. I also contributed to a scalable Learning Management System (LMS) as part of a joint project with TCS, supporting a student base of 2 lakh users across multiple universities.

On the project side, I built a Leaf Disease Detection System using a hybrid U2-Net + ResNet approach, achieving 96% classification accuracy and reducing misclassification by ~15%. I extended it into a full-stack web app (React + Flask) with image upload, bilingual results, severity scoring, and an LLM chatbot for real-time farmer queries.

I’m also passionate about end-to-end pipelines for real environments: I built a Real-Time Multi-Object Tracking Pipeline using RT-DETR-R50 and ByteTrack for Indian road environments, designed to maintain stable persistent IDs across video streams and produce annotated batch outputs with CSV logs. Finally, I developed a multilingual RAG-based guidance platform that retrieves and reranks Bhagavad Gita verses and generates grounded responses with real-time TTS across 11 regional languages, using an async FastAPI + SSE streaming backend with background warmup orchestration.