How Dashmap built an AI-powered crash detection app in 48 hours with Prolific
A team of three Imperial College graduates needed to train AI to analyze cycling crashes, fast. During the AI Engine hackathon, they used Prolific to source human judgments on accident footage and collected results in minutes.
The data helped them build Dashmap, an iPhone app that turns your phone into a cycling dashcam with AI-powered crash detection and emergency response.
The challenge: Training AI to understand cycling accidents in just 2 days
Saffan Firdaus and his two teammates from Imperial College had ambitious plans for the AI Engine hackathon. They wanted to build an app that combines navigation optimized for cyclists with dashcam recording and AI crash detection.
The concept was straightforward: your phone records while you cycle, automatically saves footage when it detects a crash, and uses AI to generate a summary that can be sent to emergency contacts or police. But making it work required solving a complex problem in less than two days.
"We wanted to understand who is at fault when a crash happens, what happens in the video," Saffan explained. "We're trying to make a summary of the video and who is at fault in the accident."
To train their AI model to analyze crash footage accurately, the team needed human judgment data. They had to show real cycling accident videos to people and collect their assessments of what happened and who was responsible. This ground truth data would train their AI to make similar judgments automatically.
The timeline was tight. With only 48 hours for the entire hackathon, they couldn't spend days recruiting participants through traditional methods. They needed quality data from real people, and they needed it instantly.
The solution: Rapid data collection through Prolific
The Dashmap team used Prolific to solve their data collection challenge. They built a survey asking participants to watch cycling crash footage and provide two key pieces of information: who was in the wrong, and a description of what happened.
With Prolific's platform, they launched their survey to the entire participant pool of 200,000+ people. The team didn't set specific filters during the hackathon, though they noted Prolific's extensive filtering capabilities for future research. "Because it was a hackathon and we had to move fast, we just chose all participants," Saffan said.
The speed of response exceeded their expectations. After sending the survey, results started flowing back within 10 minutes. The rapid turnaround meant they could collect the human judgment data they needed and move straight to training their AI model, all within the hackathon's tight deadline.
The results: From hackathon project to published app
The human feedback collected through Prolific became the foundation for training Dashmap's crash detection AI. The team used these results on top of their existing video generation AI from memories.ai to build a system that can analyze cycling accidents and generate accurate summaries.
By the end of the hackathon, Dashmap had evolved from concept to working prototype. The judges recognized the team's achievement, and the team won the ‘Best Use of Prolific’ award.
But they didn't stop there. They've since published Dashmap on the App Store for iPhone, making it available to cyclists who need protection on the road.
The team plans to add automatic emergency contact calling in future updates, using the AI-generated summaries to quickly communicate what happened and get help when cyclists need it most.
Right now, Dashmap is free to use. The team is continuing to refine their AI and expand features based on real-world cycling data.
Need to collect training data for your AI models quickly? Prolific connects you with 200,000+ verified participants who can provide quality human feedback for your models in minutes. Learn more about Prolific for AI.
