Skip to main content
Uber AI

DragonCrawl: Generative AI for High-Quality Mobile Testing

23 April / Global
Featured image for DragonCrawl: Generative AI for High-Quality Mobile Testing
Image
Fig 1: High-level overview of DragonCrawl. The image of the Dragon was generated by OpenAI’s DALL·E
Image
Fig 2: Transformer layers in DragonCrawl’s model.
Precision@1Precision@2Precision@3ParametersEmbedding size
MPNet (base)0.97230.96230.9423110M768
MPNet (large)0.97260.95270.9441340M768
T50.970.95470.933811B3584
RoBERTa0.96890.95120.946482M768
T5 (not tuned)0.92310.92130.921311B3584
Image
Fig 3: Example of how imperceptible noise can fool a Machine Learning model. This is not a hypothetical example, .
Image
Image
Figure 4: DragonCrawl going online in Brisbane, Australia after trying for 5 minutes.
Image
Figure 5: DragonCrawl restarting the app to request a trip.
Image
Figure 6: Future mobile tests as RAG applications powered by the Dragon Foundational Model (DFM)
Image
Figure 7: Model-quality fly wheel.
Juan Marcano

Juan Marcano

Juan Marcano is a Staff Machine Learning Engineer, Tech Lead, and the father of DragonCrawl. Juan works on the Core Automation Platform team and specializes in training and productionizing CV, NLP, and large language models. Juan's job is to train heroes.

Mengdie Zhang

Mengdie Zhang

Mengdie Zhang is a Software Engineer on the Core Automation Platform team working on DragonCrawl's models and DragonCrawl's holistic reliability and observability.

Ali Zamani

Ali Zamani

Ali Zamani is a Senior Software Engineer on the Core Automation Platform team focusing on backend and mobile end-to-end testing. He leads DragonCrawl's backend, CI, and other efforts to productionize DragonCrawl.

Anam Hira

Anam Hira

Anam Hira is a Machine Learning Engineer on the Core Automation Platform team working on DragonCrawl's CI and models. Anam (one of the heroes Juan has trained) developed the first version of DragonCrawl during an internship.

Posted by Juan Marcano, Mengdie Zhang, Ali Zamani, Anam Hira

Category: