Sai Vishwas

I’m a B.Tech student at IIT Kharagpur (2022–2026) focused on AI/ML engineering—turning unstructured data into reliable, production-ready experiences. I enjoy designing end-to-end pipelines, from data extraction to model-backed user interactions, with an emphasis on accuracy and scalability.

In my internships, I built an OCR and LLM extraction pipeline for insurance documents using Docling and Groq, including query optimization and benchmarking across Gemini, LLaMA, and OpenAI to improve prompt reliability. I also automated policy comparison with Excel (pandas, openpyxl), wrapped secure conversion APIs in Flask, and integrated insurance app workflows through Appian Connected Systems.

I’ve also developed RAG-based applications using Flask APIs, LangChain, Pinecone, and Gemini 2.0 Flash, including robust PDF processing with chunking and regex filtering. Beyond retrieval, I built a mental-health chatbot in Streamlit with session state, chat history, and safety-focused system prompts.

My projects extend into full-stack and agentic systems: an AI coding tutor with React/TypeScript and Express using OpenRouter-routed LLMs, a sandboxed multi-language code execution engine with Monaco Editor and SSE streaming, and PitchShark AI— a LangGraph multi-agent app delivering real-time pitch feedback using ASR with Groq Whisper. I’ve also fine-tuned LLaMA-2 for abstractive summarization with Transformers/PEFT (LoRA) and trained/evaluated models with reproducible PyTorch pipelines.