SmartPort AI is an end-to-end maritime risk intelligence system designed to predict, monitor, and act on vessel delays in congested port environments. It transforms raw AIS (Automatic Identification System) movement data into actionable operational alerts, identifying vessels at risk of exceeding the critical 120-minute berthing delay window.
Access the real-time dashboard here: https://smartport-ai-risk-early-warning.streamlit.app/
- 🖥️ Cloud Command Center (Streamlit): A high-visibility web dashboard for port authorities to monitor live vessel risk scores and operational status.
- 🤖 AI Operational Assistant (Telegram): An automated bot that delivers real-time critical alerts (🔴) and handles on-the-go queries via an asynchronous tactical interface.
- ☁️ Live Data Backbone (Google Sheets): A cloud-synchronized "Single Source of Truth" that connects the ML engine with the front-end for zero-latency data updates.
- 🧠 Predictive Core (XGBoost): A calibrated machine learning pipeline that transforms raw AIS telemetry into high-precision delay probabilities.
Data Source: Container Ship Tracking Dataset (Kaggle)
Port operations rely on tight berthing windows. Delays beyond 120 minutes have cascading economic impacts on the entire global supply chain.
SmartPort AI answers the critical operational question:
"Which vessels are likely to exceed the 120-minute delay threshold, and what is the prioritized operational response?"
- Encapsulated Pipeline: Feature engineering (vessel dynamics, time intervals) is baked directly into the Scikit-Learn pipeline object (
.pkl). - Zero-Friction Ingestion: The model safely accepts raw, unformatted AIS data, dynamically handling missing values and type casting.
- Inference: A calibrated XGBoost model classifies delay probability with high precision.
- Streamlit Cloud Dashboard: Real-time web interface for visualizing risk levels and history.
- Bidirectional Integration: Reads live data from Google Sheets and triggers tactical notifications via Telegram.
- Enterprise Security: Fully powered by Streamlit Secrets (TOML) to manage encrypted API credentials.
- Single Source of Truth: Cloud-based repository (via
gspread) for immediate operational visibility and easy team access. - Data Integrity: Every prediction includes a unique SHA-256 hash (prediction_id) for total audit traceability.
- Asynchronous Caching: Background syncing ensures zero-latency responses to user queries.
- Executive Reporting: Automatic push notifications with categorical breakdowns (🔴/🟡/🟢) and direct action protocols.
| Risk Level | Score Range | Operational Meaning | Suggested Action |
|---|---|---|---|
| 🔴 CRITICAL | > 0.80 | High likelihood of >120 min delay | Immediate intervention (reassign berth) |
| 🟡 WARNING | 0.50 – 0.80 | Elevated risk | Monitor ETA and AIS stability closely |
| 🟢 NORMAL | < 0.50 | Low risk | Routine operations |
app.py: The main Cloud Command Center (Streamlit Dashboard).01_Scripts/: Utility scripts for data cleaning and training.02_Data/: Full dataset hierarchy (Raw, Working, Validation).03_Notebooks/: End-to-end development pipeline (EDA, Modeling).04_Models/: Serialized.pklfiles for the model and pipelines.05_Outputs/: Prediction results and exported risk alerts.telegram_bot.py: The core AI Analyst engine for Telegram.logs_builder_sheets.py: Main synchronization engine for Cloud Operational Logs.requirements.txt: Optimized system dependencies.
- ML & Analytics: Python, XGBoost, Scikit-learn, Pandas.
- Web Dashboard: Streamlit, Plotly.
- Cloud & API: Google Sheets API (
gspread), OpenAI API (GPT-4o-mini). - Interface: Telegram Bot API (
python-telegram-bot). - Security: SHA-256 Hashing, Streamlit Secrets (TOML).
This project uses Streamlit Secrets for secure cloud deployment. To run locally, ensure you have a .streamlit/secrets.toml file with:
TELEGRAM_TOKEN&TELEGRAM_CHAT_ID[gcp_service_account]credentials block for Google Cloud.
Developed as part of the SmartPort AI Automation Project - 2026