GhostPilot

GhostPilot is an offline AI-powered Network Copilot developed for ISRO Hackathon PS-13. It analyzes satellite network telemetry using Machine Learning, a rule engine, and the Phi-3 Large Language Model running locally through Ollama to generate intelligent network diagnostics and future risk predictions.

---

Features

• Offline AI inference using Ollama
• Phi-3 Large Language Model
• Machine Learning-based Risk Prediction
• Future Risk Trajectory Prediction
• Rule-based Network Health Evaluation
• AI-generated Network Analysis
• Historical Report Storage
• JSON Telemetry Input

---

Tech Stack

• Python
• FastAPI
• Scikit-learn
• Ollama
• Phi-3
• NumPy
• JSON

---

Run

python app.py

---

API Endpoints

1. Health Check

GET /health

Returns the backend health status.

Response

{
  "status": "OK",
  "service": "GhostPilot AI Backend",
  "model": "phi3",
  "version": "1.0"
}

---

2. Analyze Network

POST /analyze

Analyzes satellite telemetry using the ML model and Phi-3 LLM.

Request

{
  "latency": 45,
  "packet_loss": 2,
  "cpu": 60
}

Response

{
  "status": "success",
  "network_status": "Warning",
  "predicted_risk": 55,
  "generated_at": "2026-06-28 17:20:00",
  "model": "phi3",
  "telemetry": {
    "latency": 45,
    "packet_loss": 2,
    "cpu": 60
  },
  "reason": "...",
  "possible_root_cause": "...",
  "potential_impact": "...",
  "recommended_actions": "...",
  "summary": "..."
}

---

3. Analysis History

GET /history

Returns all previously generated network analysis reports stored in the history/ folder.

---

4. Future Risk Trajectory Prediction

GET /predict/trajectory

Predicts future network risk values using a Linear Regression model trained on historical risk scores.

Response

{
  "history": [
    {
      "time": 0,
      "risk": 100
    },
    {
      "time": 1,
      "risk": 99
    }
  ],
  "prediction": [
    {
      "time": 20,
      "risk": 26
    },
    {
      "time": 21,
      "risk": 23
    }
  ]
}

Description

• Loads historical risk scores from the history/ directory.
• Trains a Linear Regression model using previous analysis results.
• Forecasts the next 10 network risk values.
• Supplies data for the Future ML Trajectory Predictions chart in the GhostPilot dashboard.

---

Project Workflow

1. Receive network telemetry.
2. Predict network risk using the ML model.
3. Evaluate network status using the rule engine.
4. Generate AI-powered diagnostics using Phi-3 via Ollama.
5. Save the analysis report.
6. Train a Linear Regression model on historical risk scores.
7. Predict future network risk trajectory for visualization.

Presentation Link: https://canva.link/vfcqv13ojqmukn5

Team Project

Team Members:

• Shruti Gupta
• Shivangi Gupta