At iQuasar, LLC, I built ProposalPro, a RAG application that retrieves relevant context from Qdrant and uses an LLM to draft proposals, with a React frontend. I also built an end-to-end outline-generation pipeline with OCR-based document ingestion, embeddings, Qdrant retrieval, and LLM generation exposed through FastAPI.
For scanned English documents, I built an ingestion path using Tesseract and OpenCV preprocessing, and evaluated EasyOCR and Docling with RapidOCR as CPU-only options. I also built Chat-With-Docs AI App for chatting with uploaded documents using Ollama and LM Studio, and Call GPTApp, a voice assistant integrating the OpenAI ChatGPT API with Twilio and Deepgram.
My backend work includes building a system-log processing and live-monitoring app with Redis and BullMQ, and leading back-end and architecture work for Project Chinar. My research at Indian Institute of Technology, Delhi included energy optimization in NUCA-based mesh SoC architecture using reinforcement learning, QUIC application classification with K-means, and parallel sorting in C++.
