AskMyDoc

• 

  Working of the smart chat assistant

• 

  Initial screen

💡 Inspiration

While navigating through dense research papers 📄, legal documents ⚖️, and lengthy technical manuals 🧠, we found ourselves endlessly skimming and scrolling 😩. The traditional PDF readers felt static and unhelpful. So, we thought — what if your documents could talk back? 🗣️ That’s how AskMyDoc was born — a smart, interactive PDF assistant 🧞‍♂️ that helps users truly understand rather than just read their documents.

🚀 What it does 💪

AskMyDoc is your AI-powered reading companion 🤖📚. Here’s what it brings to the table:

• 📤 Upload any PDF document
• 🖥️ View it in a split-screen PDF reader
• 💬 Ask questions about the whole document or specific pages
• 📄 Get context-aware, accurate answers that reference page numbers
• 🧠 Maintains a chat history linked to your reading journey
• 🧭 Like having an expert co-reader guiding you in real-time

🛠️ How we built it 🧪

• Frontend: Built with Streamlit 🧑‍🎨 for a clean and interactive split-screen interface 📑💬.
• Backend: FastAPI handles uploads, routing, and real-time responses ⚙️.
• PDF Processing: PyMuPDF extracts text from each page quickly and efficiently 🏃‍♂️.
• Embeddings: Open AI Embeddings

• Vector Database: FAISS database to store vector embeddings of the PDF.

  • Page-level and full-document embeddings created 🧩
  • Stored in vector databases for accurate semantic retrieval 🔍📦

• LLM Agent: Integrated a ReAct(Reasoning and Action)-style agent 🧠 using OpenAI APIs to answer, reason, and clarify 🔄.

• State Management: Chat history tracked 🕰️ with roles and page metadata, for a more personalized feel 👥📘.

⚔️ Challenges we ran into 🧗‍♂️

• 🧩 Balancing page-level vs full-document embeddings without leaking context.
• 🕑 Keeping latency low even with huge PDFs and long chats.
• 🎨 Designing a distraction-free UI that blends clarity and interactivity.

🏆 Accomplishments that we're proud of 🎉

• 🎯 Built an end-to-end RAG (Retrieval-Augmented Generation) pipeline.
• 📖 Developed a ReAct agent that not only answers but thinks and reasons.
• 🧼 Created a smooth, intuitive UI merging reading & real-time assistance.
• 📎 Linked chat history to document structure for seamless reference.

📚 What we learned 👨‍🎓

• How to build a full-stack LLM-powered app from scratch 🏗️.
• Deep understanding of document chunking 📄 vs global context 🌍.
• Learned to blend AI capabilities with thoughtful UX for better collaboration 👥✨.

🔮 What’s next for AskMyDoc 🚀

• 🗂️ Multi-document support: Compare and query across multiple PDFs!
• 📴 Offline mode: Use open-source LLMs locally for privacy and speed 🛡️.
• 🖊️ Interactive annotations: Get answers linked to text and export highlights!

Built With

• fastapi
• streamlit