Threat Tracker

Inspiration:

The app was born from the need to respond to global crises like the ongoing wars in Palestine, Ukraine, and Myanmar. Which have made the importance of real-time, location-based threat awareness more critical than ever. While these conflicts are often headline news, people living far from the conflict zones may lack the immediate understanding of how quickly conditions change on the ground. Our inspiration came from a desire to bridge that gap by leveraging technology to provide a solution that could offer real-time updates about dangerous areas, not just in warzones but in urban centers and conflict-prone regions around the world.

How we built it:

Our app was developed with scalability and responsiveness in mind, given the complexity of gathering real-time data from diverse sources. For the backend, we used Python to run a Reflex web app, which hosts our API endpoints and powers the data pipeline. Reflex was chosen for its ability to handle asynchronous tasks, crucial for integrating with a MongoDB database that stores a large volume of data gathered from news articles. This architecture allows us to scrape, store, and process incoming data efficiently without compromising performance.

On the frontend, we leveraged React Native to ensure cross-platform compatibility, offering users a seamless experience on both iOS and Android devices. React Native's flexibility allowed us to build a responsive interface where users can interact with the heat map, see threat levels, and access detailed news summaries all within the same app.

We also integrated Meta LLaMA, a hyperbolic transformer model, which processes the textual data we scrape from news articles. The model is designed to analyze and assess the threat level of each news piece, outputting both the geographical coordinates and a risk assessment score. This was a particularly complex part of the development process, as fine-tuning the model to provide reliable, context-aware predictions required significant iteration and testing.

Challenges we faced:

The most pressing challenge was data scraping, particularly the obstacles put in place by websites that actively work to prevent scraping. Many news websites have anti-scraping measures in place, making it difficult to gather comprehensive data. To address this, we had to get creative with our scraping methods, using dynamic techniques that could mimic human-like browsing to avoid detection.

Another major challenge was iOS integration, particularly in working with location services. iOS tends to have stricter privacy controls, which required us to implement complex authentication mechanisms and permissions handling. Additionally, deploying the backend infrastructure presented challenges in ensuring that it scaled smoothly under heavy data loads, all while maintaining low-latency responses for real-time updates.

We also faced hurdles in speech-to-text functionality, as we aim to make the app more accessible by allowing users to interact with it via voice commands. Integrating accurate, multi-language speech recognition that can handle diverse accents and conditions in real-world environments is a work in progress.

Accomplishments we're proud of:

Despite these challenges, we successfully built a dynamic heat map that allows users to visually grasp the intensity of threats in different geographical areas. The Meta LLaMA model was another major achievement, enabling us to not only scrape news articles but also analyze and assign a threat level in real time. This means that a user can look at the app, see a particular area highlighted as high risk, and read news reports with data-backed assessments. We've created something that helps people stay informed about their environment in a practical, visually intuitive way.

Moreover, building a fully functional app with both backend and frontend integration, while using cutting-edge machine learning models for threat assessment, is something we're particularly proud of. The app is capable of processing large datasets and serving actionable insights with minimal delays, which is no small feat given the technical complexity involved.

What we learned:

One of the biggest takeaways from this project was the importance of starting with the fundamentals and building a solid foundation before adding complex features. In the early stages, we focused on getting the core infrastructure right—ensuring the scraping, data pipeline, and database were robust enough to handle scaling before moving on to model integration and feature expansion. This allowed us to pivot more easily when challenges arose, such as working with real-time data or adjusting to API limitations.

We also learned a great deal about the nuances of natural language processing and machine learning, especially when it comes to applying those technologies to dynamic, unstructured news data. It’s one thing to build an AI model that processes text in a controlled environment, but real-world data is messy, often incomplete, and constantly evolving. Understanding how to fine-tune models like Meta LLaMA to give reliable assessments on current events was both challenging and incredibly rewarding.

What’s next:

Looking ahead, we plan to expand the app’s capabilities further by integrating speech-to-text functionality. This will make the app more accessible, allowing users to dictate queries or receive voice-based updates on emerging threats without having to type or navigate through screens. This feature will be particularly valuable for users who may be on the move or in situations where typing isn’t practical.

We’re also focusing on improving the accuracy and scope of our web scrapers, aiming to gather more diverse data from a broader range of news sources while adhering to ethical guidelines. This includes exploring ways to improve scraping from difficult sites and even partnering with news outlets to gain access to structured data.

Beyond these immediate goals, we see potential in scaling the app to include predictive analytics, using historical data to forecast potential danger zones before they escalate. This would help users not only react to current events but also plan ahead based on emerging patterns in conflict areas. Another exciting direction is user-driven content, allowing people to report and share information about dangerous areas directly through the app, further enriching the data landscape.

Built With

• hyperbolic
• javascript
• mongodb
• react-native
• reflex
• xcode