LLM Powered API for Fraud Detection.

Project Workflow Overview

The Fraud Detection API is designed to process financial transaction data for fraud analysis, specifically handling CSV files. The workflow comprises a series of steps that facilitate asynchronous processing and analysis.

Workflow Steps:

1. Upload Transactions:

   • Users can upload a CSV file containing transaction data via the /upload endpoint. Uploaded files are temporarily stored in the TransactionFiles directory. (To-Do: Will be migrated to an AWS storage bucket).

2. Analyze Transactions:

   • The user then initiates the analysis by calling the /analyze/:fileID endpoint, passing the file_id obtained from the upload step. The analysis process is started asynchronously. This endpoint accepts a set of fraud detection rules in the request body.

      {
         "rules": [
            "Rule 1 description",
            "Rule 2 description",
            // additional rules...
         ]
      }
   

3. Poll Analysis Status:

   • While the analysis is ongoing, the user can check its status through the /analyze/:analysisId/status endpoint. The status provides insights into whether the analysis is Processing, Complete, or has Failed.

4. Retrieve Analysis Results:

   • Once the analysis is complete, the results can be fetched from the /results/:analysisId endpoint using the provided analysisId. Additional Results may be returned as JSON files, which are stored in the ResultFiles directory. (To-Do: Will be migrated to an AWS storage bucket) If present, the files will be referenced in the API Response.

5. Health Check:

   • The API also includes a /health endpoint, allowing users to verify if the API service is operational.

How it works

Onboarding

Build the Image:

docker build -t frauddetection .

Run the Container

docker run -p 8080:8080 frauddetection

Access the application at http://localhost:8080

Swagger API Documentation

The Fraud Detection API comes with an integrated Swagger UI.

Accessing Swagger UI:

• To access the Swagger UI, first ensure the API service is running.
• Navigate to http://localhost:8080/swagger/index.html in your web browser.
• The Swagger UI page will display a list of all available API endpoints with their expected parameters and responses.
• You can interact with the API directly from this page by expanding individual endpoint details, entering required parameters, and executing requests to see the responses in real-time.

Tech Stack

Golang

• The application is built in Golang, selected for its robust support for concurrent and asynchronous processing using goroutines. This feature is crucial for efficiently handling the asynchronous tasks and large-scale data processing inherent in fraud detection.

OpenAI Assistants API

• For fraud analysis, the project utilizes the OpenAI Assistants API. This Assistants API aids in analyzing transaction data to detect potential fraud patterns.