Skip to content

cenkerozkan/LawRag-latest

BranchesTags

Repository files navigation

📌 Law-RAG

A Retrieval-Augmented Generation (RAG) project using LangChain with MongoDB for message history – This project focuses on legal documents, allowing users to select relevant legal PDFs based on their input using semantic similarity and then generate responses accordingly.

🚀 Features

✅ Semantic Similarity-Based Legal PDF Selection
✅ Integration with LangChain for RAG
✅ Efficient Retrieval and Query Processing for Legal Documents
✅ MongoDB Integration for Message History Storage

🛠️ Tech Stack

  • LLM Framework: LangChain (for retrieval and generation)
  • Embedding Models: Google
  • Vector Database: FAISS, In Memory Vector Store
  • LLM: Gemini 2.0 Flash, Gemma3 4B
  • Backend: FastAPI, Python
  • Database: MongoDB (for message history)
  • Storage: Local File System

📝 To Do

  • Implement pdf selector using semantic similarity with user input
  • Implement Document repositories for MongoDB and PostgreSQL
  • Implement MongoDB connector
  • Implement message history storage in MongoDB
  • Increase k in semantic pdf simulator for multiple rag.
  • Refactor rag_service, remove chat_thread related methods
  • Create a new service for managing chat threads.
  • Add new fields into message history part in MongoDB Model
  • Bring xml based instruction prompts
  • Implement rag service logic
  • Bring streamlit UI || FastAPI UI
  • Add metadata filtering for FAISS queries.
  • Implement a supabase watcher as an independent component.
  • Implement a new UI with streamlit.
  • In RAG service, change the parameter of send message from chat model to uuid.
  • Move FAISS to base doc repository class.
  • For each document repository, create a method that inserts related docs with uuids.
  • Implement a generic repository for custom documents uploaded by the user.
    • Implement a file upload mechanism.
    • Use InMemoryVectorStore for custom user documents.
  • Bring a scaffolding mechanism for MongoDB repositories.
  • Bring a scaffolding Mechanism for document repositories.
  • Implement a caching mechanism with redis for accessing chats
  • Use nltk to count tokens for each user message.
  • Bring more document repositories for different law areas to increase accuracy.
  • Increase the security in the prompt.
  • Create a word map for each law document to change the pdf selector examples
  • Add used_pdfs field in message model
  • Bring a more flexible solution in rag_service _pdf_selector.
  • Try vertex ai gecko multilingual embedding.

🚀 How to Run

Prerequisites

  • MongoDB running locally
  • Python 3.8+

Installation & Setup

  1. Clone the repository

    git clone https://github.com/yourusername/law-rag.git
cd law-rag

  2. Install dependencies

    pip install -r requirements.txt

  3. Start FastAPI server

    uvicorn app:app --reload --port 8000 --host 0.0.0.0

  4. Access the API at http://localhost:8000/ for test UI.

HakMateRAG Backend API Documentation

A FastAPI-based backend for the HakMateRAG application, providing services for managing chat threads, processing PDF documents, and performing RAG (Retrieval Augmented Generation) queries.

Base URL

https://your-hakmaterag-api-base-url.com

All endpoints described in this documentation should be prefixed with this base URL. For example, to access the create chat thread endpoint:

POST https://your-hakmaterag-api-base-url.com/api/chat_service/create

Table of Contents

Global Error Handling

The API implements global exception handling. All responses, whether success or error, follow the ResponseModel format:

{
  "success": true,        // boolean: Indicates if the request was successful
  "message": "Success",   // string: A human-readable message
  "data": {},             // object | null: The response data, if any
  "error": null           // string | null: An error message or code, if any
}

Common HTTP Status Codes:

  • 200 OK: The request was successful.
  • 400 Bad Request: The request was malformed or contained invalid parameters.
  • 401 Unauthorized: Authentication is required and has failed or has not yet been provided. The token is invalid or missing.
  • 422 Unprocessable Entity: The request was well-formed but was unable to be followed due to semantic errors (FastAPI default for validation errors).
  • 500 Internal Server Error: An unexpected error occurred on the server.

All protected endpoints require an Authorization header with a Bearer token: Authorization: Bearer <your_jwt_token>

Chat Thread Service API Endpoints

Base path: /api/chat_service

Create Chat Thread

Creates a new chat thread for a user.

  • Endpoint: POST /api/chat_service/create
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Request Body: 
    {
  "chat_name": "string",         // Name of the chat thread
  "user_id": "string",           // UUID of the user
  "anonymous_user_id": "string"  // Optional: UUID of the anonymous user
}
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "Chat thread created successfully",
  "data": {
    "chat": {
      "chat_name": "My Legal Questions",
      "chat_id": "uuid-generated-for-chat",
      "user_id": "user-uuid-123",
      "anonymous_user_id": "anon-uuid-456",
      "created_at": "2025-05-21T14:30:00Z",
      "updated_at": "2025-05-21T14:30:00Z",
      "pdf_content": [],
      "history": []
    }
  },
  "error": null
}
  • Error Response (500 Internal Server Error): 
    {
  "success": false,
  "message": "Chat thread creation failed",
  "data": {},
  "error": "" // Potentially more specific error from service
}

Delete Chat Thread

Deletes a specific chat thread by its ID.

  • Endpoint: DELETE /api/chat_service/delete/{chat_id}
  • Path Parameter:
    • chat_id (string, required): The ID of the chat thread to delete.
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "Chat thread deleted successfully",
  "data": {},
  "error": null
}
  • Error Response (500 Internal Server Error): 
    {
  "success": false,
  "message": "Chat thread deletion failed",
  "data": {},
  "error": "" // Potentially "Chat with id: {chat_id} not found" if service returns more detail
}

Get All Chat Threads

Retrieves all chat threads for a given user ID.

  • Endpoint: GET /api/chat_service/get_all_chat_threads/{user_id}
  • Path Parameter:
    • user_id (string, required): The ID of the user whose chat threads are to be retrieved.
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "Sohbetler başarıyla alındı", // "Chats retrieved successfully"
  "data": {
    "threads": [
      {
        "chat_name": "Legal Chat 1",
        "chat_id": "uuid-chat-1",
        "user_id": "user-uuid-123",
        "anonymous_user_id": null,
        "created_at": "2025-05-20T10:00:00Z",
        "updated_at": "2025-05-20T11:00:00Z"
        // "history" is excluded
      },
      {
        "chat_name": "Another Case",
        "chat_id": "uuid-chat-2",
        "user_id": "user-uuid-123",
        "anonymous_user_id": null,
        "created_at": "2025-05-21T09:00:00Z",
        "updated_at": "2025-05-21T09:30:00Z"
      }
    ]
  },
  "error": null
}
  • Error Response (500 Internal Server Error): 
    {
  "success": false,
  "message": "Chat thread retrieval failed",
  "data": {},
  "error": ""
}

Get Chat History

Retrieves the message history for a specific chat thread.

  • Endpoint: GET /api/chat_service/get_chat_history/{chat_id}
  • Path Parameter:
    • chat_id (string, required): The ID of the chat thread.
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "Chat history retrieved successfully",
  "data": {
    "history": [
      {
        "created_at": "2025-05-21T14:30:05Z",
        "role": "user",
        "content": "Hello, can you help me with a legal question?",
        "web_sources": null
      },
      {
        "created_at": "2025-05-21T14:30:10Z",
        "role": "ai",
        "content": "Yes, I can. Please state your question.",
        "web_sources": null
      }
    ]
  },
  "error": null
}
  • Error Response (500 Internal Server Error): 
    {
  "success": false,
  "message": "Chat history retrieval failed",
  "data": {},
  "error": ""
}

Delete All Chat Histories

Deletes all chat threads and their histories for a specific user.

  • Endpoint: DELETE /api/chat_service/delete_all_chat_histories/{user_id}
  • Path Parameter:
    • user_id (string, required): The ID of the user whose chat histories are to be deleted.
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "Chat histories deleted successfully",
  "data": {},
  "error": null
}
  • Error Response (500 Internal Server Error): 
    {
  "success": false,
  "message": "Chat history deletion failed",
  "data": {},
  "error": ""
}

Update Chat Name

Updates the name of a specific chat thread.

  • Endpoint: PATCH /api/chat_service/update_chat_name/{chat_id}/{new_chat_name}
  • Path Parameters:
    • chat_id (string, required): The ID of the chat thread to update.
    • new_chat_name (string, required): The new name for the chat thread.
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "Chat thread updated successfully",
  "data": {
    "chat": {
      "chat_name": "Updated Legal Query",
      "chat_id": "uuid-chat-1",
      "user_id": "user-uuid-123",
      "anonymous_user_id": null,
      "created_at": "2025-05-20T10:00:00Z",
      "updated_at": "2025-05-21T15:00:00Z", // Reflects update time
      "pdf_content": [],
      "history": [ /* ... existing history ... */ ]
    }
  },
  "error": null
}
  • Error Response (500 Internal Server Error):
    • If chat thread retrieval fails: 
      {
  "success": false,
  "message": "Chat thread retrieval failed",
  "data": {},
  "error": ""
}
    • If chat thread update fails: 
      {
  "success": false,
  "message": "Chat thread update failed",
  "data": {},
  "error": ""
}

Generate Chat Name

Generates an AI-based chat name from a user query using Gemini models.

  • Endpoint: GET /api/chat_service/generate_chat_name/{user_query}
  • Path Parameter:
    • user_query (string, required): The user’s initial input or question to be used for generating the chat name.
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "Chat name generated successfully",
  "data": {
    "chat_name": "My Legal Issue"
  },
  "error": null
}
  • Error Response (500 Internal Server Error): 
    {
  "success": false,
  "message": "Chat name generation failed",
  "data": {},
  "error": "Gemini API error details or internal error message"
}

Upload PDF

Uploads a PDF file and associates it with a chat thread.

  • Endpoint: POST /api/chat_service/upload_pdf/{chat_id}

  • Path Parameter:

    • chat_id (string, required): The ID of the chat thread to associate the PDF with.

  • Headers:

    • Authorization: Bearer <your_jwt_token>

  • Request (multipart/form-data):

    • file: The PDF file to upload (application/pdf)
    • file_name: The name of the file (string)

  • Success Response (200 OK):

    {
  "success": true,
  "message": "Pdf uploaded successfully",
  "data": {},
  "error": ""
}

  • Error Responses:

    • If the file is not a PDF: 
      {
  "success": false,
  "message": "Only upload PDF files please",
  "data": {},
  "error": ""
}
    • If the PDF file is too large: 
      {
  "success": false,
  "message": "PDF file is too large!",
  "data": {},
  "error": ""
}
    • If the chat thread already has 3 PDFs: 
      {
  "success": false,
  "message": "Chat thread pdf upload limit exceeded. Only 3 pdfs are allowed",
  "data": {},
  "error": ""
}
    • If the PDF content cannot be extracted: 
      {
  "success": false,
  "message": "Failed to extract text",
  "data": {},
  "error": "Extraction error details"
}
    • If the chat thread cannot be updated after upload: 
      {
  "success": false,
  "message": "Pdf content update failed",
  "data": {},
  "error": ""
}

Internal Service API Endpoints

Base path: /api/internal

Process PDF

Processes a PDF document by downloading it, extracting text, and associating it with a chat thread.

  • Endpoint: POST /api/internal/process_pdf
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Request Body: 
    {
  "conversation_id": "string", // The chat_id to associate the PDF with
  "document_id": "string",     // A unique ID for the document
  "file_path": "string",       // Path to the PDF file in storage (e.g., Supabase path)
  "file_name": "string",       // Original name of the PDF file
  "file_type": "string"        // MIME type of the file (e.g., "application/pdf")
}
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "PDF başarıyla işlendi.", // "PDF processed successfully."
  "data": {
    "extracted_text": "The full extracted text content from the PDF..."
  },
  "error": null
}
  • Error Response (500 Internal Server Error): 
    {
  "success": false,
  "message": "PDF işlenirken hata oluştu.", // "Error occurred while processing PDF."
  "data": {},
  "error": ""
}
    (This can be due to download failure, text extraction failure, or database update failure.)

RAG Service API Endpoints

Base path: /api/rag

Query RAG

Submits a query to the RAG service for a specific chat thread. The service retrieves relevant document context, optionally performs a web search, and generates a response using an AI model.

  • Endpoint: POST /api/rag/query
  • Headers:
    • Authorization: Bearer <your_jwt_token>
  • Request Body: 
    {
  "chat_id": "string",    // ID of the chat thread
  "user_id": "string",    // ID of the user making the query (Note: currently seems to be used for logging/future use, chat retrieval is by chat_id)
  "query": "string",      // The user's query
  "web_search": false   // Optional boolean (default: false): Whether to perform a web search
}
  • Success Response (200 OK): 
    {
  "success": true,
  "message": "RAG query processed successfully",
  "data": {
    "response": "The AI-generated answer based on retrieved documents and/or web search."
  },
  "error": null
}
  • Error Responses:
    • If retrieve_chat_thread fails (500 Internal Server Error): 
      {
  "success": false,
  "message": "Sohbet getirilirken bir sorun oluştu.", // "An issue occurred while fetching the chat."
  "data": {},
  "error": ""
}
    • If retrieved chat data is not in the expected format (500 Internal Server Error): 
      {
  "success": false,
  "message": "Bilinmeyen bir hata oluştu!", // "An unknown error occurred!"
  "data": {},
  "error": ""
}
    • If AI model content generation fails (from rag_service.run): 
      {
  "success": false,
  "message": "Sistemde yaşanan bir aksaklık sebebiyle şu an size yardımcı olamıyorum.", // "I cannot assist you right now due to a system issue."
  "data": null, // Or some other structure depending on the error
  "error": "Details of the Gemini API error or other internal error"
}
    • If updating the chat thread after successful generation fails multiple times: 
      {
  "success": false,
  "message": "Sohbet sırasında bir şeyler ters gitti!.", // "Something went wrong during the chat!"
  "data": {
    "response": "Şu anda size yardımcı olamıyorum." // "I cannot help you right now."
  },
  "error": "" // Potentially more specific error
}

Query with Streaming

Streams RAG responses in real time using Server-Sent Events (SSE), sending the chat_id first and then chunks of the response text in groups of three words.

  • Endpoint: POST /api/rag/query_with_streaming

  • Request Body:

    {
  "chat_id": "string",    // ID of the chat thread
  "user_id": "string",    // ID of the user making the query
  "query": "string",      // The user's query
  "web_search": false     // Optional boolean: perform a web search
}

  • Headers:

    • Authorization: Bearer <your_jwt_token>

  • Response: text/event-stream with the following message sequence:

    data: {"chat_id": "<chat_id>"}

data: {"chunk": "First three words"}

data: {"chunk": "Next three words"}

...

data: {"done": true}

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages