PortPulse

Inspiration

As Freshman Data Justice Scholars, we share a deep love for data science and a strong desire to utilize open, ethical data sources. We wanted to push ourselves out of our comfort zones and tackle a real-world problem using concepts we hadn't fully explored yet.

Our inspiration struck from observing current global conflicts, such as the closing of the Strait of Hormuz. We realized that resource price spikes often surprise consumers who are entirely unaware of the root cause—which usually begins with port congestion and shipment delays. We noticed a distinct lack of platforms displaying this crucial information concisely. Inspired by the space theme of Bitcamp, we looked to the stars and decided to use public satellite AIS (Automatic Identification System) data to build an early-warning dashboard.

What it does

PortPulse is a live supply chain dashboard designed to predict commodity price spikes days before they hit the news. The frontend features a pannable, interactive 3D globe that visualizes real-time commercial vessel movements.

With data updating every 5 seconds, the dashboard highlights "hot zones" of port congestion using color-coded indicators (where red means critical queuing and imminent delays). By spotting a spike developing at a major port like Shanghai, logistics managers and consumers can anticipate ripple effects in the global market.

How we built it

We built a full-stack application connecting real-time satellite tracking with predictive machine learning and generative AI.The Data Pipeline: Our backend, written in Python, pulls a live WebSocket feed tracking every commercial vessel on Earth. We evaluate the speed of ships near each port against a threshold to detect preliminary congestion.The Math: We calculate a live congestion metric using the following formula:$$C = \bar{S}{6h} + \bar{t}{dwell}$$Where $C$ represents the total congestion score, $\bar{S}{6h}$ is the average number of ships in the port per 6-hour time bucket, and $\bar{t}{dwell}$ is the average dwell time (how long each ship remains docked).Predictive Modeling: This congestion data is fed into an XGBoost model to predict potential price increases for the resources carried by those delayed ships.GenAI Alerts: Finally, an open-source LLM—Llama 3, running locally on our machine—reads the predictive scores and translates the data into plain-English alerts for the user.

Challenges we ran into

As freshmen, diving into advanced data science implementations was daunting. We had to quickly learn how to manage live WebSocket data streams, integrate machine learning models (XGBoost) into a real-time pipeline, and host a local LLM efficiently. Rendering a smooth, interactive 3D globe with constantly updating data points also required significant optimization and troubleshooting.

Accomplishments that we're proud of

We are incredibly proud of how well we translated our initial imaginations and draft works into a tangible, high-functioning reality. Building a tool that seamlessly integrates live satellite data, predictive AI, and a 3D user interface as freshmen is a massive milestone for us. More importantly, we proved that open, ethical data sources can produce genuinely powerful tools.

What we learned

We learned the practical realities of handling large-scale, real-time datasets and how global supply chains are intimately tied to localized port logistics. We also discovered the immense value of stepping into uncomfortable fields; struggling through the integration of Llama 3 and XGBoost taught us more about system architecture than a textbook ever could.