🧠 Semantica

Semantica is a lightweight semantic search engine for PDF documents. It processes PDF files, converts them into vectorized text chunks, and enables intelligent retrieval using embedding-based similarity search — all without using an LLM.

---

🚀 Features

• 📄 Upload any PDF file
• ✂️ Automatic chunking of document content
• 🔢 Embedding with HuggingFace (MiniLM)
• 🧠 Vector search with Qdrant
• ⚡ Fast and local — no OpenAI API required
• 📆 Built with FastAPI, LangChain, and Qdrant

---

🖥️ Tech Stack

Layer	Tool
Backend	FastAPI
Parsing	pymupdf4llm
Chunking	LangChain MarkdownTextSplitter
Embedding	sentence-transformers/all-MiniLM-L6-v2
Vector DB	Qdrant via Docker

---

🔧 Setup & Run

1. Clone the repository

git clone https://github.com/yourname/semantica.git
cd semantica

2. Install dependencies

pip install -r requirements.txt

3. Run Qdrant locally (Docker)

docker run -p 6333:6333 qdrant/qdrant

4. Start the FastAPI server

fastapi dev main

Then open the Swagger UI at: 📍 http://localhost:8000/docs

---

🦪 API Endpoints

POST /upload

Uploads and parses a PDF file. Chunks it and saves to Qdrant with embeddings.

POST /search

Send a semantic query and receive relevant chunks. Example request:

{
  "query": "Does this PDF mention 'fun' keyword?"
}

Example response:

[
  {
    "score": 0.92,
    "text": "This is a simple PDF file. Fun fun fun.",
    "source_file": "sample.pdf",
    "chunk_id": 1
  }
]

---

📜 Future Plans

• LLM-based answer generation
• Multi-document support
• Frontend interface for document search (possibly separated)

---

🤝 Contributing

Pull requests, feedback and ideas are always welcome. If you use this project, feel free to ⭐️ the repo and share your feedback.

---

📄 License

MIT