AskMyPaper: Your Research Assistant

AskMyPaper is an AI-powered research assistant that helps users interact intelligently with academic papers. Built using Streamlit and integrated with advanced models and APIs, it allows you to upload a PDF and ask questions directly about the content. It can even fetch and analyze referenced papers automatically.

---

🔍 What is AskMyPaper?

AskMyPaper bridges the gap between dense research papers and user questions.
Instead of reading an entire paper, users can upload it and ask specific questions. The app reads the document, understands the content, and provides human-like responses.

Core Features:

• Upload a research paper PDF
• Ask natural language questions
• Receive contextual answers
• Automatically extracts and downloads cited references (DOI or arXiv)

---

⚙️ Core Components

AskMyPaper is built on five key technical components:

1. Text Extraction & Chunking – Converts the PDF into readable, manageable text chunks.
2. Embeddings (HuggingFace) – Transforms text into high-dimensional vectors for similarity search.
3. FAISS Vector Store – Efficient storage and fast retrieval of similar text chunks.
4. Google Gemini LLM – Generates intelligent, context-aware answers.
5. Reference Downloaders – Automatically fetches cited papers using DOI or arXiv links.

---

🤖 HuggingFace Embeddings

Model Used: sentence-transformers/all-MiniLM-L6-v2

• Converts chunks of text into numerical vectors (embeddings)
• Captures semantic meaning to enable context-aware search
• Enables similarity search to find relevant information for user queries

---

🧠 Google Gemini LLM

Model Used: gemini-2.5-flash-preview-04-17

• A powerful large language model from Google
• Takes your question and retrieved document context to generate accurate responses
• Handles both document-specific and general research questions

---

🔑 Google Gemini API

• Connects to Gemini via the GOOGLE_API_KEY
• The application sends the user’s question and context
• Gemini returns the generated answer which is displayed in the app

---

📥 Reference Downloaders (DOI & arXiv)

AskMyPaper enhances its knowledge by reading cited papers:

DOI (Digital Object Identifier)

• Queried via: https://doi.org/
• Adds Accept: application/pdf header to retrieve the paper
• May redirect through services like Crossref or Unpaywall

arXiv

• Constructs direct URL:

• Downloads open-access papers directly

---

🔁 Workflow: From PDF to Answer

[User Uploads PDF] 
      ↓
[Text Extracted using PyPDF2]
      ↓
[References Detected via Regex]
      ↓
[Reference PDFs Downloaded (DOI/arXiv)]
      ↓
[Text Chunked using RecursiveCharacterTextSplitter]
      ↓
[Embeddings Generated via HuggingFace]
      ↓
[Stored in FAISS Vector Store]
      ↓
[User Asks Question]
      ↓
[Similar Chunks Retrieved]
      ↓
[Question + Context Sent to Gemini API]
      ↓
[Answer Generated & Displayed in Streamlit]

📌 Key Takeaways

• Contextual Understanding: Embeddings + vector search ensure highly relevant retrieval
• Intelligent Responses: Gemini synthesizes information to provide detailed answers
• Expanded Research: Automatically includes references from cited works
• Easy to Use: Streamlit interface makes everything accessible with minimal setup

---

🛠️ Tech Stack

• Frontend: Streamlit
• PDF Handling: PyPDF2
• Text Processing: LangChain, RecursiveCharacterTextSplitter
• Embeddings: HuggingFace (all-MiniLM-L6-v2)
• Vector DB: FAISS
• LLM API: Google Gemini via google.generativeai
• Reference Downloads: requests with DOI & arXiv parsing

---

📚 Future Improvements

• Add multi-PDF comparison
• Enable PDF export of chat history
• Integrate citation-based summarization
• UI enhancements with better citation linking

---

🧑‍💻 Developed By

• 22071A6651 – Sathvik Deekonda
• 22071A6662 – Urlana Suresh Kumar
• 22071A6664 – Yasaswi Kode

Final Year B.Tech Students – CSE (AI & ML), VNRVJIET (2022–2026)

---

💬 Questions or Suggestions?

Feel free to open an issue or pull request!
Let’s make research smarter and more interactive together. ✨