# InterviewSystem

**Repository Path**: zeron2022/InterviewSystem

## Basic Information

- **Project Name**: InterviewSystem
- **Description**: 基于大模型的医疗问答系统
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 1
- **Created**: 2025-08-08
- **Last Updated**: 2025-08-08

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Interview Process and Q&A System

Edited by CHENYAN WU && Ying-XW

## Setup
- Clone repository 
```
git clone https://github.com/asnbby/InterviewSystem.git
```
- Setup conda environment
```
conda create -n struct_llm python=3.9
conda activate struct_llm
```
- Install packages with a setup file
```
bash setup.sh
```
- Install pytorch module

if you are linux or windows:

```
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
```

if you are macos :

```
pip3 install torch torchvision torchaudio
```
- Download Embedding Model From Hugging-Face(We use stella_en_400M_v5 as our embedding and retrieve model)
```
git lfs clone https://hf-mirror.com/dunzhang/stella_en_400M_v5
```


## Before Use

- Make sure you have got your api-key and your api-key has been named as "api_key_${YourModelName}.txt" and saved at ./api_key/ 

- The same make sure your base_url has been named as "url_${YourModelName}.txt" and saved at ./openai_url/

## Run System
- Start Chromadb service with code
```
chroma run --path /YourPath/chromadb/SentenceBERT/interview --port 8000
```
- Start system
```
bash run.sh
```
## Acknowledgement
```
This project is intended for non-commercial use
If you want to use it for commercial use, please contact wuchenyan0823@gmail.com
```
## Technical Report (The Approach section of the report)

### Work Introduction

  In this section, I will briefly introduce the technical aspects of the project. Our main contributions are summarized as follows:
  - 1.We use Self-prompt Extraction to capture potential semantics from the interview corpus and use this method to transform the source text into context, question answer pairs, and summaries for retrieval.
  - 2.We use the LLM-based Reranker module, leveraging the semantic understanding ability of LLM to rank context, question answer pairs, and summaries in a hierarchical manner, and ultimately fusion these ranking to obtain the final retrieval results
  - 3.Finally, we propose a Chatbot with CoT module that fully utilizes Reranker's ranking results and multi-level semantic guidance for LLM to answer questions.

### Self-prompt Extraction
  Self-prompt Exraction