semantix

A portable semantic search tool for comprehending local markdown, docx, and pdf files.

Languages and Tools:

• Python
• FastAPI
• Docker
• ChromaDB
• LangChain
• HTML
• CSS
• Javascript
• Transformers (HuggingFace)

Architecture Diagram

graph TD
    subgraph Client_Side [User Interface Container]
        UI[HTML/JS Frontend]
    end

    subgraph Backend_API [FastAPI Container]
        API[FastAPI Router]
        Ingestor[Ingestion Logic: PyMuPDF/Docx]
        Splitter[LangChain Splitter]
        Embedder[Sentence-Transformers Model]
    end

    subgraph Storage_Layer [Vector Database Container]
        DB[(ChromaDB)]
    end

    UI -- "1. Upload File (POST /ingest)" --> API
    API --> Ingestor --> Splitter --> Embedder
    Embedder -- "2. Store Vectors + Metadata" --> DB

    UI -- "3. Search Query (GET /search)" --> API
    API -- "4. Embed Query" --> Embedder
    Embedder -- "5. Semantic Comparison" --> DB
    DB -- "6. Return Top-K Results" --> API
    API -- "7. JSON Response" --> UI

Loading

Installation

To build this application you simply need to clone this repo locally or you can run curl github.com/eonloop/semantix

This container comes bundled as a docker compose package, so once you've downloaded the repo run the following commands

cd semantix
docker compose up --build

Usage:

Once the application is up and running You can then access the application from

localhost:8000

Here you will be able to upload individual documents once you've uploaded the documents there is no current ability to clear the database, this will be coming in a future improvement.

Once you have uploaded your documentation you should be able to query it by typing you query into the "Enter a search query" text field and clicking search.

This should return the top 2 results from your documentation that you have uploaded into the database.