K-Pop Business Analysis Chat Bot

AI-powered chatbot for analyzing K-POP industry financial and business data using Vector DB and local LLM.

Language: English | 한글 (Korean)

Overview

This chatbot leverages Ollama's local LLM (Llama 3.1) and ChromaDB vector database to analyze K-POP industry financial data and business information. Unlike prompt-based approaches, it uses Vector DB similarity scores to objectively evaluate answer accuracy.

Features

• Local LLM-based: Ollama + Llama 3.1 with no cloud dependency
• Vector DB Search: ChromaDB for efficient document retrieval
• Accuracy Scoring: Automatic reliability calculation based on similarity (0-100%)
• Multiple Confidence Levels: Very High / High / Medium / Low / Very Low
• Streamlit Web UI: Intuitive and user-friendly interface
• Chat History: Automatic conversation logging
• Source References: Document citations for all answers

Tech Stack

Component	Technology
LLM	Ollama + Llama 3.1
Embedding	Sentence Transformers (all-MiniLM-L6-v2)
Vector DB	ChromaDB
Web UI	Streamlit
Language	Python 3.9+

Requirements

• Python 3.9+
• Ollama (Install: https://ollama.ai)
• 4GB+ RAM

Installation & Usage

1. Activate Virtual Environment

source kpopBot_venv/bin/activate

2. Start Ollama Server

OLLAMA_HOST=127.0.0.1:11435 ollama serve

Install required model:

ollama pull llama3.1

3. Run Web UI (Recommended)

streamlit run app.py

Browser opens automatically at http://localhost:8501

Or use CLI

python chat.py

Accuracy Scoring System

Automatically calculated based on Vector DB Cosine Similarity:

Score	Confidence	Description
85%+	Very High	Highly trustworthy
70-85%	High	Reliable
55-70%	Medium	Moderately reliable
40-55%	Low	Low confidence
<40%	Very Low	Not trustworthy

Project Structure

KpopBot/
├── app.py                 # Streamlit web UI
├── chat.py               # CLI chatbot
├── fileProcess.py        # PDF processing and vector DB creation
├── Data/                 # K-POP company financial documents (PDF)
│   ├── 2021년/
│   ├── 2022년/
│   ├── 2023년/
│   ├── 2024년/
│   └── 2025년/
├── vector_db/            # ChromaDB storage
└── README.md

File Descriptions

app.py

• Streamlit-based web interface
• Visual accuracy metrics display
• Detailed analysis view (per-document similarity)
• Chat history management

chat.py

• Terminal-based CLI interface
• Text-based accuracy score output
• Automatic Q&A loop

fileProcess.py

• PDF document loading and preprocessing
• Text chunking
• Vector embedding and ChromaDB storage
• Metadata management (year, quarter, company)

Usage Examples

Sample Questions:

• "What was HYBE's revenue in 2024?"
• "What is SM Entertainment's quarterly operating profit trend?"
• "What are JYP Entertainment's main revenue sources?"

Response Includes:

1. Detailed answer in markdown format
2. Accuracy metrics
3. Per-document similarity scores
4. Source references

Configuration

Modify settings in chat.py or app.py:

# LLM settings
llm = ChatOllama(
    model="llama3.1",
    base_url="http://127.0.0.1:11435",  # Ollama address
    temperature=0  # 0: accurate, 1: creative
)

# Search options
retriever=vector_db.as_retriever(search_kwargs={"k": 5})  # Top 5 documents

Troubleshooting

Q: "Connection refused" error

Ollama server not started
$ OLLAMA_HOST=127.0.0.1:11435 ollama serve

Q: "Vector DB not found"

vector_db folder missing
$ python fileProcess.py  # Process data to create vector DB

Q: Low accuracy score

Need more training data
- Add more PDFs to Data/ folder
- Run python fileProcess.py to re-index

License

MIT License

---

Last Updated: March 2026