HYPERLOCAL — A Real-Time Disruption Q&A

This is a submission for the Bright Data AI Web Access Hackathon

What I Built

HYPERLOCAL is a Q&A assistant designed to provide real-time answers about local disruptions such as traffic jams, public transport issues, protests, adverse weather conditions, and/or construction work. Users can ask questions about a specific city/area, and HYPERLOCAL leverages live web data to deliver up-to-date information.

The core problem it solves is the time-consuming nature of finding real-time disruption information. I'll use this tool myself because I've personally experienced this frustration when traveling. I had to check airline sites for delays then traffic reports to know what time to leave the house to avoid traffic jams/possible accidents etc.

Instead of manually browsing multiple websites, users can get relevant information through a simple Q&A interface. This is particularly useful for commuters, travelers or even event organizers who need timely updates.

Demo

• Repository: GitHub
• Live Application: HYPERLOCAL

Credentials

The app is currently password-protected for jury members (password sent to Noah@BrightData in order to avoid seeing my OpenAI API balance being drained out faster than expected!).

How I Used Bright Data's Infrastructure

HYPERLOCAL integrates Bright Data's MCP server primarily to enable the AI agent to perform the four key actions outlined in the challenge:

1. Discover: When a user asks about disruptions in a specific location, the agent uses tools provided by the MCP server to find relevant web pages containing real-time information. This include latest news report, traffic alert systems, public transport agency websites, and/or social media feeds.
2. Access: With Bright Data's Web Unlocker, the agent can navigate and access data from websites that might employ anti-bot measures, cookie banners that block content, or require complex interactions to reach the relevant information. Based on my own testing, the tool successfully retrieved complex reports from news feeds, posts on X and even informations contained in popups on interactive maps.
3. Extract: Once relevant pages are accessed, the MCP tools help the agent extract the structured data about the found disruptions, directly in markdown.
4. Interact: For websites that render content dynamically using JavaScript or require interactions (like the map popup I mentionned above), the MCP server enable the agent to interact with these pages as a human would.

Performance Improvements

I can effectively explain how Bright Data's MCP server significantly enhances my app performance by acknowledging an initial misstep in the project's architecture. My early misconception was to treat the MCP server as an API. This led to a system built around form-based data retrieval, where city selection triggered the MCP data fetching, followed by database caching and display. This failed to capitalize on the "real-time" functionality inherent in MCP servers.

By designing the project as a Q&A interface, where the agent directly uses MCP tools (as opposed to providing pre-fetched data in the system prompt like I did before), the capabilities are expanded. The agent can, based on conversational context and user input, dynamically generate its own optimized queries and select appropriate links, fully exploiting the real-time data.