The Problem
Organizations and individuals deal with large volumes of PDF documents — research papers, legal contracts, financial reports, policy manuals, product documentation — on a daily basis. Extracting specific insights from these documents is a time-consuming and inefficient process.
Key pain points include:
● Information Overload — Professionals spend hours manually reading through lengthy PDFs to find specific answers.
● No Conversational Access — Traditional PDF readers offer only search (Ctrl+F), not understanding or reasoning.
● Context Loss — When reading large documents, users lose track of prior sections, making cross-referencing difficult.
● Repetitive Work — The same questions get re-researched from the same documents repeatedly, with no memory or reuse.
● Accessibility Gap — Non-technical users cannot easily query complex documents without data or ML expertise.
There was a clear need for an intelligent solution that allows any user to have a natural conversation with their documents and receive precise, cited answers instantly.
Our Solution
PDF Mind is a smart, AI-powered chatbot that allows users to upload any PDF document and ask questions about it in plain natural language — receiving accurate, context-aware answers with source citations.
Key highlights of the solution:
● 📄 One-click PDF Upload — Users simply drag and drop any PDF into the interface.
● 🤖 GPT-4o-mini LLM — OpenAI’s fast and cost-efficient model powers the question-answering engine.
● 🔍 Semantic Search — The document is chunked and embedded into a vector space; questions are matched to the most relevant passages using similarity search.
● 🧵 Conversational Memory — The chatbot remembers the last 12 exchanges, enabling multi-turn conversations with full context.
● 📚 Source Citations — Every answer includes expandable source passages from the PDF, ensuring transparency and traceability.
● ⚡ Intelligent Caching — The vector store is cached per document hash, eliminating redundant re-processing on page reloads.
● 🎨 Premium Dark UI — A beautiful glassmorphism-style Streamlit interface that is intuitive for both technical and non-technical users.
Solution Architecture
┌─────────────────────────────────────────────────────────────┐
│ USER (Browser) │
│ Streamlit Web Application (Port 8501) │
└────────────────────────┬────────────────────────────────────┘
│
┌──────────▼──────────┐
│ PDF Upload Module │
│ (pypdf + io.BytesIO│
└──────────┬──────────┘
│
┌──────────▼──────────┐
│ Text Chunking │
│ RecursiveCharacter │
│ TextSplitter │
│ chunk=1200, overlap │
│ =200 tokens │
└──────────┬──────────┘
│
┌──────────▼──────────┐
│ Embedding Model │◄── OpenAI
│ text-embedding- │ API
│ 3-small │
└──────────┬──────────┘
│
┌──────────▼──────────┐
│ FAISS Vector Store │◄── @st.cache_resource
│ (Cached per PDF │ (per MD5 hash)
│ MD5 hash) │
└──────────┬──────────┘
│
┌──────────▼──────────┐
│ MMR Retriever │
│ (top-5 diverse │
│ chunks, fetch_k=10)│
└──────────┬──────────┘
│
┌──────────▼──────────┐
│ GPT-4o-mini LLM │◄── OpenAI
│ (temp=0.3) │ API
└──────────┬──────────┘
│
┌──────────▼──────────┐
│ Conversation Memory │
│ BufferWindowMemory │
│ (k=12 turns) │
└──────────┬──────────┘
│
┌──────────▼──────────┐
│ Answer + Sources │
│ → Chat UI │
└─────────────────────┘
Caching Strategy:
| Layer | Mechanism | Benefit |
| PDF Text Extraction | @st.cache_data (keyed by bytes) | No re-reading on Streamlit rerun |
| FAISS Vector Store | @st.cache_resource (keyed by MD5 hash) | No re-embedding on page reload |
| Conversation Context | ConversationBufferWindowMemory(k=12) | Last 12 Q&A turns retained in LLM context |
Deliverables
| # | Deliverable | Description |
| 1 | Working Web Application | A fully functional Streamlit app running on localhost:8501 with complete PDF chat capability |
| 2 | PDF Processing Pipeline | End-to-end pipeline: upload → extract → chunk → embed → index |
| 3 | Conversational AI Engine | GPT-4o-mini powered Q&A with sliding window memory (last 12 turns) |
| 4 | Semantic Vector Search | FAISS-based MMR retrieval returning the 5 most relevant, diverse chunks per query |
| 5 | Intelligent Cache Layer | FAISS store cached per PDF hash — zero re-embedding cost on app refresh |
| 6 | Source Citation Feature | Expandable “View source passages” panel under every AI answer |
| 7 | Premium UI/UX | Dark glassmorphism interface with animated chat bubbles, live stats, and responsive layout |
| 8 | GitHub Repository | Full source code hosted at https://github.com/shivamrawat2002/ChatWithPDF1 |
| 9 | Environment Configuration | .env support with .env.example template and .gitignore protecting secret keys |
| 10 | Documentation | Comprehensive README.md with setup instructions, architecture overview, and tech stack |
Tech Stack
| Layer | Technology | Version | Purpose |
| Frontend / UI | Streamlit | 1.35+ | Web application framework |
| LLM | OpenAI GPT-4o-mini | Latest | Natural language understanding & answer generation |
| Embeddings | OpenAI text-embedding-3-small | Latest | Semantic vector representation of text chunks |
| Vector Store | FAISS (Facebook AI Similarity Search) | 1.8+ | Fast approximate nearest-neighbor vector search |
| LLM Framework | LangGraph Classic | 1.0+ | Conversational retrieval chain & memory management |
| PDF Parsing | pypdf | 4.2+ | Extract text from PDF pages |
| Text Splitting | langchain-text-splitters | 1.1+ | Recursive character-aware document chunking |
| Caching | Streamlit Cache (@st.cache_resource, @st.cache_data) | Built-in | Zero-cost re-use of embeddings & vector stores |
| Environment | python-dotenv | 1.0+ | Secure API key management via .env |
| Tokenization | tiktoken | 0.7+ | Accurate token counting for chunking |
| Language | Python | 3.10+ | Core programming language |
Business Impact
Industries & Sectors Impacted
1. Legal & Compliance
Law firms and corporate legal teams deal with thousands of pages of contracts, case files, and regulatory documents. PDF Mind enables paralegals and lawyers to instantly query any document, reducing contract review time by an estimated 60–70% and minimizing the risk of missing critical clauses.
2. Healthcare & Pharmaceuticals
Medical researchers and clinicians can rapidly extract insights from clinical trial reports, drug documentation, and research papers. This accelerates evidence-based decision-making and reduces the time from research to actionable knowledge.
3. Financial Services
Analysts can query annual reports, prospectuses, and audit documents conversationally. Rather than spending hours reading 200-page filings, a question like “What was the net profit margin in Q3?” returns an instant, cited answer.
4. Education & Research
Students and academics can interact with textbooks, journal papers, and thesis documents. This democratizes access to complex academic content, reduces study time, and supports deeper comprehension through follow-up questioning.
5. Government & Public Sector
Policy documents, tenders, and compliance manuals can be made accessible to non-technical staff through natural language querying — improving efficiency and reducing dependency on specialized knowledge workers.
6. Enterprise Knowledge Management
Internal product manuals, SOPs (Standard Operating Procedures), and HR policy documents can be converted into interactive, queryable knowledge bases — reducing onboarding time and improving employee self-service.
Quantified Impact Potential
| Metric | Estimated Impact |
| Document review time | Reduced by 60–75% |
| Repeated information lookup | Eliminated via cached memory |
| Dependency on SMEs for document queries | Reduced by 40–50% |
| Onboarding time for document-heavy roles | Reduced by 30–40% |
| Cost of embedding re-computation | Zero (FAISS cache per document) |
Key Value Propositions
● Speed — Answers in seconds instead of hours of manual reading
● Accuracy — Grounded answers with source passage citations (no hallucination risk)
● Scalability — FAISS caching ensures performance doesn’t degrade with repeated usage
● Accessibility — Zero ML expertise required; any user can interact via natural language
● Cost Efficiency — GPT-4o-mini is OpenAI’s most cost-effective production model
Github Link : https://github.com/shivamrawat2002/ChatWithPDF1
Demo Video
