Vsuraj Kumar

At Automaton AI Infosystem, I build production-grade GenAI applications in Python, including agentic systems and multi-agent workflows integrated with AWS Bedrock and Azure OpenAI.

I designed RAG pipelines with text chunking, metadata tagging, embeddings, and hybrid search for enterprise semantic search. I also iterated on prompts and reranking to improve response consistency and engineered agents for data extraction workflows that reduced manual processing effort by an estimated 60%.

At A21.AI, I developed Model Context Protocol servers for document-processing workflows and built automated PDF extraction pipelines using LangGraph, Azure OpenAI, PyTesseract, and PyMuPDF.

In my AI Search Engine project, I developed multilingual semantic retrieval using Llama 3 on AWS Bedrock, Cohere embeddings, and Redis Vector Database. I implemented semantic and exact caching that reduced latency from about four seconds to two, and deployed the system on AWS EC2 across UAT and production environments.