Compass AI

A CoPilot-based agentic system that identifies patterns, routes escalations, learns from feedback, and gives engineers clarity instead of chaos. This repository contains the codebase for the full-stack application for viewing and analyzing internal support escalations.

Project website: https://compass-ai.my.canva.site/

Project Structure

single_source_of_truth_monorepo/
├── frontend/          # React frontend application
├── backend/           # Express.js backend API
├── package.json       # Root package.json with convenience scripts
└── README.md          # This file

System Architecture

Prerequisites

• Node.js 16+ (recommended 18+)
• npm 8+

Check versions:

node -v
npm -v

Setup

Install all dependencies

From the root directory:

npm run install:all

Or install individually:

cd frontend && npm install && cd ..
cd backend && npm install && cd ..

Running the Application

Option 1: Run from root (recommended)

Start frontend:

npm run start:frontend

• Opens http://localhost:3000

Start backend:

npm run start:backend

• Typically runs on http://localhost:3001 or http://localhost:3000 (check backend config)

Option 2: Run individually

Frontend:

cd frontend
npm start

Backend:

cd backend
npm start

Data Pipeline

The production dataset comes from the scripts in data_preprocessing/:

1. Run csv_to_json.py to convert cleaned_incidents.csv into incidents.json (raw structured data).
2. Run enrich_incidents.py to add routing metadata, confidence scores, context, and timelines. This writes incidents_enriched.json.
3. The backend automatically loads data_preprocessing/incidents_enriched.json (or any file pointed to by ESCALATIONS_DATA_PATH) at startup, so the React app always works with the latest enriched data. A fallback copy in backend/data/escalations.json is only used if the pipeline output is missing.

Both scripts require Python 3.9+ plus the packages listed in the notebooks (scikit-learn, numpy, etc.). Re-run them whenever the CSV changes or you retrain the ML model (model.pkl / label_encoder.pkl).

Development

• Frontend: React app with components in frontend/src/components/
• Backend: Express.js API in backend/
• Data: Generated via the pipeline above; no mock data is used in the UI.

Build

Build the frontend for production:

npm run build:frontend

Testing

Run frontend tests:

npm run test:frontend

Notes

• Set REACT_APP_API_URL if your backend runs on a non-default host/port.
• Override the backend dataset path with ESCALATIONS_DATA_PATH if you want to point at a different JSON file.
• CORS is enabled in the backend to allow the React frontend to fetch data.