Sajjan Singh

At Zangoh, I build multilingual OCR and document-layout pipelines that extract structured content while preserving reading order. I also develop multi-agent workflows with FastAPI, LangGraph, and LangChain.

I compare DETR-based and VLM layout approaches using mAP, and evaluate OCR recognition quality with CER and WER. I use OmniDocBench scores and task requirements to help select VLMs.

During my SDE GenAI internship at Zangoh, I built FastAPI services and agentic workflows with tool use and context handoffs. I also prepared OCR datasets and trained and evaluated models; my contributions supported the team’s Stage 3 shortlist in the NCIIPC Startup India AI Grand Challenge.

In my projects, I built an agentic RAG research assistant and a conversational RAG agent, where Redis caching and API optimization reduced average response latency by approximately 30%. I also developed a Java 17 and Spring Boot service-management application with REST APIs, JWT authentication, and role-based access.