PDFChatbot
Multimodal RAG & Vision OCR Intelligence Engine.
Constructed a robust multimodal RAG architecture combining LangChain, Google Gemini 2.5 Flash, and FAISS to facilitate precise contextual document retrieval with page-level attribution.Architected an advanced multimodal vision OCR pipeline to seamlessly extract unstructured information, intricate tables, and high-resolution diagrams from image-heavy documents.Optimized real-time performance by implementing FastAPI and Server-Sent Events (SSE) for asynchronous token streaming, significantly reducing perceived response latency.
Multimodal Document Intelligence
Dense documents such as academic research and technical manuals frequently include tables and complex vector diagrams. By pairing PyMuPDF with vision models and vector embeddings in FAISS, the system extracts tabular data while citing specific page origins.
Real-Time Streaming Retrieval
By streaming answer generation over Server-Sent Events (SSE) via FastAPI, users receive instantaneous token feedback as the model derives answers with pinpoint context.