StraitWatch

Inspiration

Global trade relies on a few narrow maritime points such as the Strait of Hormuz, where a large portion of the world’s oil supply passes each day. When disruptions occur, whether from geopolitical tension, tanker incidents, or environmental hazards, the consequences can ripple through energy markets and global supply chains, worst of all: cause chaos and panic.

However, early warning signals of these disruptions are often scattered across different sources: news reports, vessel movement data, and environmental alerts. We wanted to explore whether these signals could be combined into a single system that detects disruption risk earlier and helps explain what is happening in real time.

StraitWatch was inspired by the idea of building an AI-assisted maritime intelligence tool that can monitor these signals continuously and translate them into clear, actionable insights.

We wanted to make it friendly, combining a Bloomberg terminal in a layman way. Which is why we also incorporated (via solana) fake paper money which helps beginner investors get into investing with AI analytics and information (such that is provided via alphaseeks and etc).

What it does

StraitWatch is an AI-powered system that monitors critical shipping chokepoints and detects early signs of disruption.

The platform analyzes signals from multiple sources including:

global news events
tanker movement and vessel activity
environmental and weather conditions

From these inputs, StraitWatch generates:

a real-time crisis risk score
a shipping disruption score
a prediction of near-term oil price movement
an automated briefing explaining what signals triggered the alert

The system also includes pattern recognition that compares current conditions to known disruption patterns and historical maritime crises. In addition, users can run scenario simulations to see how disruptions might impact risk levels and oil markets.

Together, these tools turn scattered signals into a clearer picture of what is happening now and what could happen next.

How we built it

StraitWatch was built as a multi-layered maritime intelligence system that combines live signal ingestion, structured event understanding, predictive modeling, and an interactive dashboard.

Data ingestion and signal collection: We ingest global news and event data to monitor disruptions around major shipping chokepoints such as the Strait of Hormuz. We also incorporate vessel activity, environmental conditions, and supporting market signals to build a broader view of maritime risk. These signals are continuously collected and normalized into a shared pipeline.

Structured event extraction with Gemini: Raw articles and event snippets are processed with the Gemini API to convert unstructured text into structured records. Gemini helps identify event type, affected region, severity, supply risk, and other useful metadata, allowing us to move from noisy headlines to machine-readable intelligence.

Storage and live updates with MongoDB: We use MongoDB as the central data layer for storing raw events, vessel records, engineered features, predictions, historical patterns, and simulation outputs. This gave us a flexible way to work with evolving schemas during the hackathon and made it easier to serve live dashboard updates.

Feature engineering and risk modeling: Signals from news events, vessel behavior, and environmental alerts are aggregated into rolling feature vectors that describe the state of a shipping chokepoint over time. These features power our crisis risk score, shipping disruption score, and near-term oil impact forecast.

Pattern recognition and scenario simulation: We added a pattern recognition layer that compares current conditions against historical disruption profiles and known behavioral templates, helping identify whether a live situation resembles a prior crisis. On top of that, we built a scenario simulation system that lets users test escalation events such as transit reduction, rerouting pressure, or partial chokepoint closure and immediately see the projected impact on risk and markets.

Presage for predictive simulation support: We incorporated Presage as part of the project’s predictive intelligence layer to support simulation-style reasoning around disruption scenarios. This helped strengthen the system’s ability to move beyond static monitoring and toward forward-looking risk exploration.

Voice briefings with ElevenLabs: To make the output more immediate and useful in a live operations context, we used ElevenLabs to generate short spoken briefings. This allows StraitWatch to summarize what changed, why the alert fired, and what the projected impact is in a format that feels closer to a real-time intelligence briefing.

Audit trail with Solana: We used Solana to create a lightweight immutable record of finalized crisis reports and model outputs. By anchoring alert summaries and prediction timestamps on-chain, we added a verifiable audit trail for when a report was generated and what the system predicted at that time.

Design and interface workflow with Figma: We used Figma to plan the dashboard layout, interaction flow, and visual hierarchy before implementation. This helped us design a cleaner intelligence-style interface for maps, timelines, event feeds, and simulation panels without losing clarity during development.

Frontend dashboard: The final system is presented through a live dashboard that visualizes vessel activity, risk timelines, event feeds, historical pattern matches, and scenario simulation results in one place. The goal was to create a tool that not only detects disruption, but also explains and stress-tests it.

Challenges we ran into

One of the biggest challenges we encountered was integrating multiple heterogeneous data sources into a single coherent pipeline. The system relies on signals from news intelligence, vessel movement data, environmental alerts, and market indicators. Each of these sources arrives in very different formats and time frequencies. News feeds might update every few minutes, vessel telemetry may arrive as dense geospatial coordinates, and environmental alerts follow a completely different structure.

Because of this, building a unified feature pipeline required careful normalization and transformation. We needed a way to convert these diverse inputs into consistent feature vectors that could describe the state of a maritime chokepoint over time. This meant designing intermediate schemas that could handle text-based signals, geospatial vessel data, and structured alert information without losing useful context. Ensuring that this pipeline remained stable while data continuously updated was one of the more technically complex parts of the project.

Another challenge was event understanding and noise filtering. News data in particular can be extremely noisy. Many articles may reference similar keywords without actually indicating a meaningful disruption. We used the Gemini API to convert raw articles into structured event records, but we still had to carefully design prompts and validation steps so the extracted signals remained useful for modeling. Finding the right balance between capturing relevant geopolitical signals and filtering out irrelevant news required experimentation during development.

Building the pattern recognition layer also presented challenges. Historical maritime disruptions rarely follow identical patterns, so defining useful reference templates required combining domain intuition with empirical experimentation. We created disruption profiles based on typical signals seen in past crises, then implemented similarity scoring to compare these templates against the live feature vector.

The scenario simulation system introduced another layer of complexity. Rather than simply displaying predictions from live data, we wanted users to explore hypothetical disruptions such as partial chokepoint closure or increased tanker rerouting. To support this, we designed a mechanism that perturbs the system’s feature inputs and reruns the prediction models in real time. Ensuring that these simulations behaved realistically without introducing unstable model behavior required careful parameter tuning.

Despite these challenges, working through them helped us better understand how complex global systems can be translated into structured signals and predictive insights.

Accomplishments that we're proud of

Building a system that integrates news intelligence, shipping data, and environmental signals into one monitoring platform
Implementing pattern recognition to identify similarities with historical maritime disruptions
Creating an interactive scenario simulation tool to test potential escalation scenarios
Designing a clear dashboard that turns complex data streams into understandable insights

What we learned

Building StraitWatch taught us a lot about how difficult it is to transform scattered global signals into actionable intelligence. One of our biggest takeaways was that real-world monitoring systems rarely rely on a single data source. News reports, shipping activity, environmental alerts, and market signals all provide partial views of a situation. The real challenge is combining those signals into a unified representation that can meaningfully describe what is happening in a complex system.

We also learned how important feature engineering and signal design are in predictive systems. While it can be tempting to rely heavily on large models, much of the project’s effectiveness came from carefully designing interpretable features such as vessel speed changes, reroute counts, and event severity scores. These features made it possible not only to generate predictions but also to explain why those predictions were made.

Working with structured event extraction was another major learning experience. Using Gemini to convert raw news text into structured signals highlighted how powerful large language models can be when used as part of a data pipeline rather than as the entire solution. By turning unstructured articles into machine-readable data, we were able to integrate news intelligence directly into the modeling process.

We also gained insight into the challenges of working with geospatial and maritime data. Vessel movement data is noisy and often incomplete, and meaningful signals must be extracted from patterns such as traffic density, speed changes, and rerouting behavior. This reinforced how important domain-specific reasoning is when designing anomaly detection systems.

Another important lesson was the value of explainability and interpretability. In systems that deal with geopolitical risk or market predictions, users need to understand why an alert was triggered. Designing the system so that it can clearly identify the signals driving each prediction made the platform much more useful and credible.

We also explored how technologies like blockchain and voice synthesis can play supporting roles in intelligence systems. Anchoring prediction summaries on Solana demonstrated how immutable records can provide transparency and auditability, while integrating ElevenLabs allowed us to experiment with delivering intelligence briefings in a more natural format.

Finally, building this project reinforced how important scoping and prioritization are in a hackathon setting. Our initial ideas were much broader, but narrowing the focus to a single chokepoint and a clear set of capabilities allowed us to build a more coherent and functional prototype. That constraint ultimately helped us produce a system that feels more like a real operational tool rather than just a technical experiment.

Overall, the project helped us better understand how AI, geospatial data, and event intelligence can work together to monitor and anticipate disruptions in complex global systems.

What's next for StraitWatch

Future improvements could include expanding the system to monitor additional global chokepoints, integrating satellite imagery and more advanced vessel anomaly detection, and refining market prediction models. Ultimately, StraitWatch could evolve into a broader platform for monitoring and anticipating global trade disruptions.