VESSL Llama RAG Risk Assessment

This repository contains source code for a research project conducted by SNUCEM focused on developing risk assessment models for LLM-based RAG (Retrieval-Augmented Generation) systems.

Notice

⚠️ This is a private repository. Copying, distributing, or using this code without explicit permission from the authors is prohibited.

For inquiries about this project, please contact @pikaybh or below.

Todos

• Frontend 구현 (Select할 수 있는 Input List를 각 회사의 API로 구현)
• 모델 Troubleshooting 구현
• [-] Loader 추상화
• 한국어 모델(EEVE) Resource 구현
• Query 처리 구현 (잘 안되면 raw json으로 처리)

Getting Started

Using Vessl CLI

Install the VESSL CLI this command:

pip install vessl

Set up the VESSL CLI with this command:

vessl configure

Create a run by sepcifying the YAML configuration file:

vessl run create -f run.yaml

For beginners, a simple “Hello World” example is recommended.

Running the server

To run the server, execute the following command:

$ cd app
$ uvicorn main:app --reload

API

This app provides a REST API.

GET /health
POST /v1/openai/kras/invoke

API Endpoints

GET /health
    Returns the current status of the RAG system
    
    Response 200:
    {
        "status": "OK",
        "version": "1.0.0"
    }

POST /v1/openai/kras/invoke
    Performs risk assessment on the given input
    
    Request Header:
    {
        "Authorization": "X-API-KEY <API_KEY>"
    }

    Request Body:
    {
        "query": "string",      // 이거 아님! TODO: 수정 필요
        "context": "string",    // 이거 아님! TODO: 수정 필요
        "response": "string"    // 이거 아님! TODO: 수정 필요
    }

    Response 200:
    {
        ...
    }

Architecture

The architecture of this project is as follows:

app
├── assets
│   ├── ...
├── models
│   ├── ...
아, 귀찮다... 직관적으로 이름 지어놨으니깐, 알잘딱깔쎈 하셈

이하 필요하다고 생각되는 내용 있으시면 알아서 추가해주세요... 어차피 예쁘게 정리해줘도 안 읽을거고, 결국 코딩 내가 다 하게 될 거잖아 ㅋ

Contact

Byunghee Yoo:
- Email: [email protected]
- Page: https://pikaybh.github.io/
- Github: https://github.com/pikaybh

Copyright

© 2025 SNUCEM. All Rights Reserved.