Skip to content
Sharib Khan
← All work

Productivity Platform

A RAG-powered chatbot that turns any PDF into a conversation, with streaming responses, persistent chat history, auto-generated follow-up questions, and an image-to-chat mode for visual content.

Summarize section 2Show key figures
Fig. 02 · PDF ChatBot system illustration

01Context

A solo project built for anyone who needs to pull structured insight out of unstructured documents, from students to professionals and researchers.

02Problem

Reading a long PDF to find one answer is slow, and generic chatbots can't ground their answers in a specific document.

03Approach

Chunk and embed the uploaded PDF, retrieve the relevant chunks per question through a RAG pipeline, and stream the answer back through an LLM. Includes an image-to-chat mode so visual content in the document isn't lost.

04Core Experience

Built the upload flow, streaming chat UI and auto-generated follow-up questions, so there's always a next question to ask instead of a blank input.

05AI Pipeline

Chunked and embedded documents into Pinecone, retrieved relevant passages per query, and used the Groq API for low-latency streamed generation, including responses grounded in extracted image content.

06Challenges

Keeping retrieval relevant on long, unevenly structured PDFs, and streaming partial tokens to the client without breaking markdown formatting mid-stream.

07Decisions

Generated follow-up questions from the same retrieved context as the answer, rather than a separate prompt, so suggestions stayed grounded in what the document actually said.

Layer 01 / 4

Next.js

Handles the PDF upload and renders the streaming chat UI along with suggested follow-up questions.

Key features

  • Streaming responses
  • RAG over uploaded PDFs
  • Chat history
  • Auto-generated follow-up questions
  • Image-to-chat
  • Vector search

Result

Shipped end-to-end solo, from PDF ingestion through the RAG pipeline to a streaming chat interface. Image-to-chat was added after the base RAG flow was working.

What I'd improve

Add multi-document chat to compare across several PDFs in one conversation, and citation highlighting back to the source page.

Next dossier · 03

Ecommerce Website