Real-time Analytics
Grab needed a more accurate and efficient way to map fast-changing Southeast Asian road networks, where manual intervention and conventional mapping coverage were insufficient for local traffic rules and complex geometries.
Reported impact
Grab improved lane and sign localization accuracy, reduced manual mapping work and operational costs, and created a more trustworthy, hyperlocal map platform that better serves users, driver-partners, and enterprise customers.
Grab uses GPT-4o vision fine-tuning inside GrabMaps to automate detailed mapmaking from street-level imagery gathered across Southeast Asia. The company collects millions of 360-degree images through motorbike drivers and pedestrian partners, then combines that imagery with map tiles to train models that can localize speed-limit signs, turn restrictions, road geometries, lane dividers, and other hyperlocal features. An initial speed-limit-sign workflow improved from 67% to 80% accuracy after only two rounds of fine-tuning on 100 sample cases, and the same approach helped the model reason through elevated roads, occlusions, and context that had previously required manual correction. The resulting location-intelligence data supports Grab’s ride-hailing and delivery operations and is also packaged for enterprise customers, while related AI work is expanding into accessibility and support assistants for broader platform use.
Street-level imagery from drivers and pedestrian partners is combined with map tiles and sent through GPT-4o fine-tuned vision workflows. The resulting sign, lane, and road-geometry annotations feed GrabMaps, which then supports internal mobility services and external enterprise location-intelligence offerings.
"GPT‑4o vision fine-tuning enables the service to correctly localize traffic signs and count lane dividers"
"Lane count accuracy increased by 20%"
"Speed limit sign localization improved by 13%"
"Starting with a baseline accuracy of 67%, the model improved to 80% after two rounds of fine-tuning"
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
OpenAI Sample Oct 8, 2026 Axios AI is already a huge part of how Axios Local works. At the core, what we’re trying to do is prove that you can run a sustainable, profitable local news model that delivers high-quality journalism to every community in… Axiomizer · custom GPTs OpenAI Sample Modified · Oct 7, 2026 SensoRy AI SensoRy AI needed to convert complex wildfire sensor readings into clear, immediate guidance for responders because detecting a fire is only useful if firefighters and officials can understand the situation quickly… answers in seconds ChatGPT · ChatGPT Voice 
MongoDB Sample Archived · Sep 30, 2026 Amadeus IT Group The shift to New Distribution Capability (NDC) bypassed traditional GDS buffers, sending massive search traffic directly to airlines’ systems. Furthermore, MongoDB's distributed architecture and sharding capabilities enable Amadeus to scale horizontally,… MongoDB Atlas · MongoDB Enterprise Advanced 