50x Faster AI and 7x Faster Training: Optimizing Bing Maps’ Deep Learning
Bing Maps QAS
Microsoft Bing is one of the leading global web service providers, leveraging artificial intelligence and machine learning in a suite of their products, including in search, generative AI, and machine translation.
Users can also use Microsoft’s web mapping service, Bing Maps , to perform location-based searches for navigational purposes. This complex functionality relies on a process called geocoding, which is powered by machine learning. Unfortunately, geocoding results aren’t always accurate and can lead users astray.
The Challenge
Query Annotation System (QAS) for Precise Geocoding
Bing approached Janea Systems to develop a query annotation system (QAS) for automating the collection and processing of user-submitted geocoding errors in Bing Maps. Over time, Janea’s role expanded to optimizing Bing Maps’ deep learning inference pipeline for enhanced accuracy and efficiency.
Client impact
Janea Systems’ involvement began with a simple automation task and evolved into full-fledged machine learning optimization and inference performance enhancements. In order to achieve this result, our team:
Automated Query Processing
- Developed a PowerShell-based automation script to extract and process Discrepant Satisfaction (DSAT) reports, eliminating the need for manual review and improving model regression tests.
- Optimized pipeline execution to automate query corrections and feed them back into Bing’s databases.
Optimized Bing Maps' Deep Learning Model
- Refactored TensorFlow and PyTorch implementations in DeepCAL, improving training and inference efficiency.
- Converted large single files into smaller TFRecord files to leverage parallel I/O and improve batch processing.
- Reworked a deep training algorithm from scratch in line with Python best practices for enhanced data handling.
- Optimized the TensorFlow pipeline by replacing Keras API with tf.data API and aligning with HuggingFace standards.
By addressing Bing’s complex engineering challenges, Janea delivered measurable improvements in automation, scalability, and model efficiency, reinforcing its expertise in AI-driven search optimization and large-scale ML engineering.
If you’d like to explore how your organization can automate processes or optimize machine learning models, feel free to contact us. We're here to help you navigate these complexities and uncover new opportunities for your business.
Other Client Successes
Let's talk about your project
Let's talk about your project
Ready to discuss your software engineering needs with our team of experts?