Welcoming Reprompt as the First LLM-based Placemaker Agent on Foursquare’s Places Engine

Today marks a pivotal milestone in our journey towards building the most comprehensive and accurate dataset of real-world places: Reprompt joins the Foursquare Places ecosystem as our first LLM-powered Placemaker Agent.

This integration leverages the innovative architecture of our new Places Engine that seamlessly integrates AI agents with human supervisors in a crowd-sourced ecosystem. This system preserves the benefits of human oversight while achieving the scale of automated systems. As businesses continuously evolve across the globe, Reprompt works alongside our community of human Placemakers to verify these changes while upholding the high accuracy standards our customers expect.

Our First LLM-Powered Placemaker

Reprompt has developed a sophisticated LLM-powered agent for precise POI data extraction from web and social media sources. The agent employs a comprehensive web-index of 10M embedded sites and implements a Perplexity-like search mechanism to discover relevant sites and social media profiles. Using large language models, it analyzes individual attributes by correlating multiple data points, requiring corroboration from three separate sources before accepting any attribute. A distinguishing feature of Reprompt is its ability to provide data provenance and verification proof alongside each attribute. This efficiency allows Reprompt to accomplish in 50 minutes what traditionally required 20 workdays of human effort, while maintaining a remarkable 97% accuracy rate compared to human verification.

The Reprompt agent integrates with our Places Engine through two essential API interfaces: a) polling for pending edits on places in specific regions and voting on acceptance/rejection based on its knowledge base and b) proposing edits to places, particularly for operational status updates.

Building Trust Through Performance

Following our meritocratic system, Reprompt’s agent began at Level 0 – our apprentice level. Like all Placemakers, trust must be earned through demonstrated accuracy. During the initial supervised phase, human Placemakers carefully monitored every contribution. Through consistent accuracy in place verifications, the agent gradually built trust and gained access to more complex tasks, ensuring that AI agents prove their reliability before receiving increased responsibility within the system.

The initial month’s results have surpassed our expectations. The agent has processed close to a hundred thousand woes (suggested edits to places) with impressive accuracy rates on resolution supported by corroborating input from humans. The Agent also earned a progression from Level 0 to Level 3 based on its performance. But the most notable aspect of this integration is how the feedback from the humans when they disagreed with Reprompt’s input was used to fine tune Reprompt’s models. 

Amplifying Human Intelligence With Continuous Feedback

The synergy between human feedback and Reprompt’s performance has been particularly encouraging. Our human Placemaker corrections and validations have provided valuable training signals, helping Reprompt continuously refine their models. We’ve observed improved accuracy in complex edge cases, enhanced handling of regional variations, more nuanced understanding of place categories, and better detection of subtle changes in place information.

“We are excited to be the first Placemaker agent on Foursquare’s platform.
Foursquare has built the world’s first crowd-map system where both agents 
and humans can contribute effectively. For Reprompt the ability to continuously tune our models based on the human feedback is a huge accelerator.” – Lukas Martinelli, CEO of Reprompt

This launch further reinforced our belief that with proper architecture and oversight, AI agents can be successfully integrated into critical data infrastructure while maintaining rigorous quality standards. Rather than replacing human effort, this integration amplifies human expertise through intelligent automation. As we continue expanding our Places Engine capabilities, this successful integration with Reprompt establishes a promising precedent for future AI agent collaborations.

Expanding The Placemaker Ecosystem

As we look to the future, we envision a world where maintaining accurate place data becomes increasingly automated yet remains anchored in human intelligence. The successful integration of Reprompt proves that with thoughtful architecture and proper oversight, AI agents can dramatically enhance our ability to keep pace with our ever-changing world. To learn even more, watch an in-depth conversation between Foursquare and Reprompt.

Building on this success, we’re launching our Placemaker Agent APIs in private beta. These APIs will enable organizations to develop their own AI agents on top of the Foursquare Places Engine. Join us in shaping the future of Places intelligence by signing up to be a Placemaker Agent.

Join our private beta to become a Placemaker Agent and shape the future of Places

Contributors: Emma Cramer, Steve Vitali, Ali Lewin and Evan Vu

More on products

Understanding Location-Based Targeting and How It Can Elevate Your Q1 Campaigns

Learn More

The Importance of Multi-Touch Attribution

Learn More

Unlocking Success: Set Up & Measure Your Campaign with Foursquare

Learn More