Skip to content

README.md

Latest commit

 

History

History
195 lines (119 loc) · 9.6 KB

README.md

File metadata and controls

195 lines (119 loc) · 9.6 KB

FinMycelium

A comprehensive Multi-Agent Financial Event Reconstruction platform powered by AI

中文版 | English

Python License Status LangGraph

Financial Event Reconstruction • Multi-Agent • Intelligent • Modular • AI-Powered

How can the complete chronological process of a specific event, particularly a financial event, be reconstructed as a structured timeline such as a Gantt chart from large-scale, heterogeneous, and noisy real-world data?

FinMycelium, a name borrowed from "finance" and "mycelium" is a Financial Event Reconstruction Platform that reconstructs the complete financial event process as a structured timeline from multi-source, diverse public documents. Our platform is built on a large model–based multi-agent system, in which agents cooperate to collect, match, and summarize large-scale, heterogeneous, and noisy real-world data, ultimately building a comprehensive and structured reconstruction of the event.

Finance Mycelium (菌丝体)：一个如菌丝体般在异构金融事件碎片中自主延伸、连接并重构真相的动态网络.

Features

Core Capabilities

  • Multi-Source Data Collection: Gather relevant data from diverse sources, including news articles, social media, official reports, and other public documents.
  • Data Matching and Summarization: Identify, align, and condense useful information across heterogeneous sources to directly support financial event reconstruction.
  • Event Reconstruction: Leverage a large model–based multi-agent system to autonomously integrate, connect, and reconstruct coherent event narratives from fragmented financial data.
  • Structured Presentation: Visualize the reconstructed timeline using Gantt charts and Event Developmental Sequence to deliver a clear, structured overview of the event.
  • Web Interface: Provide an intuitive, Streamlit-based interactive UI for real-time financial event analysis, exploration, and visualization.

To Do List

  • Address the "exceeding length limits" issue in both prompts and generations.
  • Re-architect FinMycelium as a fully agent-based platform.
  • Optimize the event reconstruction pipeline to improve processing speed.
  • Expand the range of data sources for data collection.
  • Implement multi-language support in the web interface.

Project Demonstration

12.25.mp4

Quick Start

Prerequisites

  • Python 3.11+
  • API Keys for LLM services (OpenAI, DeepSeek, etc.)

Installation

git clone https://github.com/AgenticFinLab/FinMycelium.git
cd FinMycelium
pip install -r requirements.txt
cp .env.example .env

Edit .env with your API keys and configuration settings. FinMycelium relies on the lmbase repository for core LLM inference and utility components.

Basic Usage

Using the Pipeline

from finmy.pipeline import FinmyPipeline
import yaml

# Load configuration
with open("configs/pipline.yml", "r") as f:
    config = yaml.safe_load(f)

# Initialize and run pipeline
pipeline = FinmyPipeline(config)
pipeline.lm_build_pipeline_main(
    data_sources=[
      "https://edition.cnn.com/2025/11/11/uk/zhimin-qian-cryptocurrency-fraud-scheme-jailed-uk-intl-hnk",
      "https://www.theguardian.com/uk-news/2025/nov/11/fraudster-who-hid-in-london-is-jailed-over-bitcoin-scam",
      "https://www.cps.gov.uk/cps/news/two-people-imprisoned-their-key-roles-largescale-money-laundering-case"
    ],
    query_text="What is the case involving fraud and money laundering by Qian Zhimin?",
    key_words=["fraud", " money laundering"]
)

Web Interface

streamlit run examples/utest/test_web_interface.py

Detailed Guidance

Reconstructed Events

  • 📦 Reconstructed Result: FinalEventCascade.json reconstructed event details.

  • 🗓️ Gantt Chart Visualization: (FinalEventCascade_gantt.html) timeline visualization of FinalEventCascade.json.

    Double-click the .html file for visualization on your browser.

  • 📊 Event Cascade Data: (Class_Build_Event_Cascade_*.json) reconstructed Event Developmental Sequence.

  1. Lan Tian Ge Rui Fraud Case (天津蓝天格锐特大非法集资案):
Builder Type Result
AgentEventBuilder 📦 Reconstructed Result
🗓️ Gantt Chart Visualization
ClassEventBuilder 📊 Event Developmental Sequence
  1. Hainan Real Estate Foam (海南房地产泡沫):
Builder Type Result
AgentEventBuilder 📦 Reconstructed Result
🗓️ Gantt Chart Visualization
ClassEventBuilder 📊 Event Developmental Sequence
  1. Tulip Bubble (郁金香泡沫):
Builder Type Result
ClassEventBuilder 📊 Event Developmental Sequence
  1. FTX Crypto Exchange Collapse (交易所FTX崩盘):
Builder Type Result
AgentEventBuilder 📦 Reconstructed Result
🗓️ Gantt Chart Visualization
ClassEventBuilder 📊 Event Developmental Sequence

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

License

Acknowledgments

We would like to thank the following projects and communities:

  • LangGraph - Multi-agent orchestration
  • Streamlit - Web interface framework
  • Bettafish - Media crawler framework
  • LlamaIndex - Indexing and retrieval
  • Various LLM providers and search APIs for their excellent services

Contact & Support

Please feel free to open an issue if you encounter any problems or have any suggestions.