YKS AI RAG

A RAG (Retrieval Augmented Generation) based assistant for question solving and topic explanation, developed for YKS (Higher Education Institutions Examination) students.

Features

• Local Document Processing: Processes lecture notes and books in PDF and TXT formats.
• Vector Search: Finds relevant content, formulas, and sample questions quickly using FAISS.
• Smart Solution: Generates step-by-step, understandable mathematical solutions using Google Gemini API (Gemini 2.5).
• Fast and Lightweight: Does not require a GPU, can run on CPU.

Installation

1. Install Requirements

   pip install -r requirements.txt

2. Set Environment Variables Copy .env.example to .env and add your Gemini API key.

   copy .env.example .env

   Open .env and enter your GEMINI_API_KEY.

   It is also recommended to add HF_TOKEN (Hugging Face Token) (https://huggingface.co/settings/tokens):

   HF_TOKEN=hf_...
   

Usage

1. Adding and Indexing Documents

Place your PDF or TXT files in the documents/ folder. A sample file konu_anlatimi_ornek.txt is included.

To start indexing:

python -m ingest.ingest_documents

This process splits documents into chunks, generates embeddings, and saves them to the index/ folder.

2. Starting the Backend

Start the API server:

uvicorn app.main:app --reload

The server will run at http://localhost:8000.

3. Asking Questions

You can ask questions by sending a POST request while the API is running.

Example Request (Curl):

curl -X POST "http://localhost:8000/ask" -H "Content-Type: application/json" -d "{\"question\": \"What are the roots of x^2 + 5x + 6 = 0?\"}"

Example Request (Python):

import requests

url = "http://localhost:8000/ask"
payload = {"question": "What are the roots of x^2 + 5x + 6 = 0?"}
response = requests.post(url, json=payload)
print(response.json())

Project Structure

• app/: Main application code (API, Core, Utils)
• ingest/: Document processing scripts
• documents/: Source documents
• index/: Vector database files