Fin_care is a web application that allows users to manage their financial data and insights by multiple AI driven functionalities.

README_zh.md

RESTful API and event-driven architecture

git clone https://github.com/CSUYSD/Anti-Scam-Financial-Management-Assistant.git

cd Anti-Scam-Financial-Management-Assistant/Backend
./setup.sh
cd ../Frontend
npm install

3.1 API keys Put your api key in the following format in IDEA run configuration OPENAI_API_KEY="your openai api key" ALPHA_VANTAGE_API_KEY="your alpha vantage api key"

if you want enable AWS S3 for file persistence, follow these steps:

1. Configure your own aws credentials
2. create a bucket
3. replace the text in Backend/src/main/resources/application.properties with your own bucket name and region(see the following picture)

run the following command to clone the wiseflow repository

git clone https://github.com/TeamWiseFlow/wiseflow.git
cd wiseflow
cp .env.example .env
nvim .env

then you will see the following picture, replace the text with your own wise flow api key notice: If you find your .env file has any variable that not present in the picture, delete it. till this step, you don't have pocketbase db account yet, so you need to create one, run

docker compose up

notice: first time running the container will encounter an expected error, because you haven't created an admin account for the pb repository yet. please keep the container running, open http://127.0.0.1:8090/_/ in your browser, and create an admin account according to the prompt(you must use email), then fill the admin email and password into the .env file, and restart the container.

cd ../Backend
./start.sh
cd ../Frontend
npm start

A cluster of three multi-functional agents—RecordRetriever, StockAnalyzer, and WeatherAnalyzer—integrated with a chatbot, operates using a three-level decision tree for autonomous function selection based on user prompts. Here’s an overview of the decision-making flow:

• Level 1: Agent Selection
  The chatbot initially selects the relevant agent based on the type of user query. Each agent (e.g., StockAnalyzer, RecordRetriever) is chosen according to its defined capabilities.

• Level 2: Function Selection within Agent
  The chosen agent then identifies the specific function based on the user’s prompt. For example, if StockAnalyzer is activated, it will determine if the request is related to stock data, company details, or historical trends.

• Level 3: Response Verification and Optimization
  The response generated by the agent is verified to ensure it aligns with the user’s query. If necessary, further optimizations are applied to refine the output.

The Anti-scam Transaction Agent is a real-time monitoring tool designed to detect suspicious activities in uploaded transactions. It leverages RabbitMQ for message queuing, WebSocket for real-time alerts, and Spring AI for behavior analysis, providing efficient, real-time processing.

• Implementation Logic
  This agent functions through an event-driven approach. When a client submits a transaction via a POST request, the system asynchronously sends the transaction data through RabbitMQ to the analysis agent. The agent reviews transaction history from the last 10 days as context, evaluating the current transaction using pre-configured behavioral models in Spring AI.

• Suspicious Activity Detection
  If suspicious behavior is identified, the agent generates a warning message (prefixed with “WARNING”) and sends it instantly to the frontend via WebSocket. The frontend then decides how to manage these warnings, ensuring flexible handling.

This chatbot component integrates Spring AI with the Chroma Vector Database to provide an advanced conversational experience, enhanced by Retrieval-Augmented Generation (RAG) and contextual memory.

• File Vectorization and Storage
  Uploaded files are automatically vectorized and stored in the vector database, enabling efficient retrieval and knowledge augmentation during conversations.

• RAG-Enhanced Responses
  When RAG is enabled in a client request, the chatbot utilizes the vectorized content to generate enriched, knowledge-based responses.

• Context Memory Management
  The system employs ChatMemory and MessageChatMemoryAdvisor to maintain context, using unique conversationIds for each chat session. This ensures isolated context and chat memories for separate conversation threads.

This feature includes a standalone interface that retrieves up to the 20 most recent transactions from the user’s database records. Through a highly customized prompt manager, these transactions are converted into a detailed prompt for the LLM (Large Language Model). The LLM then references stored documents in the vector database as context to generate a personalized, recent financial status report for the user.

This module allows users to customize and retrieve specific web content using the Wiseflow Engine. It provides full management, display, and synchronization of investment themes and articles.

• Features
  Users can add, edit, or delete themes and articles. Each theme includes a name and description, while each article stores the title, URL, summary, and content. Data is fetched through a database interface and displayed at page initialization.

• Soft Delete and Real-Time Sync
  Themes can be soft-deleted by marking their status as false, allowing for easy restoration. The module supports real-time synchronization and search, subscribing to data changes and automatically refreshing the page to display the latest themes and articles. Keyword filtering is also supported, enabling users to quickly find specific content.

1. Stock Price

• Real-time Stock Price driven by AlphaVantage API

2. Chat History

• Support user to view their chat history

3. File Persistence

• Support user to upload their financial data and files to the AWS S3

• Use Redis to cache frequently accessed data
• Use async processing to improve performance

• Low coupling and high reusability code.

• Private files(e.g. bank statement) stored in AWS S3 are protected by ACL
• User authentication and authorization by using Spring Security 6(integrated with JWT)

• Responsive design by using Tailwind CSS
• event-driven architecture for smooth user experience by using WebSocket and RabbitMQ
• Intuitive UI/UX
• Google map API for location-based service
• Real-time Stock Price driven by AlphaVantage API

• User management
• Transaction management
• File management
• Chat history management

• Implement a mobile application
• Implement a distributed storage system on TiDB
• Chatbot memory management