At Nest By Nepra, I built Birbal, a domain-specific GenAI chatbot that handles business queries with agentic workflows, FAISS-powered retrieval, conversational context, and voice interaction. It achieved over 95% uptime.
I also architected an EPR-OCR platform using LLaMA-based models for document classification and data extraction, reducing manual processing time by over 40%. I fine-tuned LLaMA models with PEFT / LoRA and deployed real-time inference APIs on AWS.
Previously, at Solution Analysts, I developed document-understanding pipelines combining OCR, NLP, CNN, and XGBoost. On my Sketch2Code personal project, I built a multimodal pipeline that converts hand-drawn wireframes into HTML/CSS, with a live demo and documentation published on GitHub.

