Bing engineers were manually reviewing and updating geolocation error reports to feed them back to the geocoding model.

Manual Report Processing

The TensorFlow implementation in Bing Maps' DeepCAL model was 10x slower than PyTorch.

Deep Learning Bottlenecks

Fixed encoding length in TensorFlow's implementation increased file sizes by 3-4x, which slowed batch processing.

Batching Limitations

A training algorithm in an internal modeling tool was suboptimal.

Algorithm Inefficiency

The TensorFlow input pipeline worked slower than its PyTorch counterpart.

TensorFlow Latency

Microsoft

We eliminated manual query correction efforts by automating Bing’s geocoding query processing pipeline.

100% Automated Error Correction

TensorFlow implementation became 50x faster than its original state.

50x faster TensorFlow

Breaking up large files resulted in enhanced efficiency on 2 GPUs.

2x speedup in batch processing

The underperforming algorithm was reworked for 7x acceleration.

7x faster training runs

TensorFlow and PyTorch input pipelines reached performance parity, with TensorFlow gaining +30% speed on dual CPU.

30% speedup on dual GPU

50x Faster AI and 7x Faster Training: Optimizing Bing Maps’ Deep Learning

medium_Bing Logo.png

small_Bing Logo.png

thumbnail_Bing Logo.png

Bing Logo.png

OtoNexus

Implemented QR code encoding and decoding for efficient data collection and retrieval. 

Patient Data Scanning

Integrated accelerometer to detect device movement and enable automated system responses. 

Device Motion Detection 

Optimized real-time analysis within the embedded C++ environment. 

On-Device Data Analysis 

Designed efficient power management protocols to extend battery life through automated standby mode and optimized charging cycles. 

Battery Management 

Developed state machines for transitions between charging, low battery, and standby modes. 

State Machines 

Enabled clear and responsive state displays, providing intuitive feedback during device operations. 

GUI Improvements 

João

Lead Software Engineer

Branimir

Senior Software Engineer

Gokce 

Janea Systems helped accelerate clinical assessment tool from prototype to production by implementing real-time data processing and battery management.

On-Device Analysis and Data Management for OtoNexus

On-Device Analysis & Data Management for Ear Infection Assessment

Migrate and adapt an enterprise platform to a client-managed environment. 

Infrastructure Migration 

Transition from Okta SSO to Ping Federate with a complete authentication logic rewrite. 

Authentication System Revamp 

Upgrading & refactoring an inherited mix of JS, TypeScript, CoffeeScript, Python, and Terraform codebase. 

Technical Debt Management 

Fixing limitations in Signals Notebook that affect data processing speed. 

Functionality Gaps  

Full CFR 21 Part 11 and GXP compliance within less than a year, enabling regulatory approval for continued use. 

GXP Compliance within a Year

Reducing reliance on Excel macros and second-person reviews by automating calculations. 

6,000+ Scientists Benefited 

Scientists gained the ability to create custom workflows using the External Data Connector and Jupyter Notebooks. 

Workflow Customization 

Enhancing the pace of pharmaceutical research by implementing scalable data processing solutions for large datasets.

R&D Acceleration

Supported the creation of a dedicated division for scientific workflow automation and facilitated large-scale data processing projects. 

Long-Term Strategic Impact 

Filip

Vedran 

Dragos 

Discover how Janea Systems facilitated R&D platform integration for a biopharmaceutical company, setting standards in scientific research and development.



R&D Platform Integration for Pharmaceutical Research

medium_Logo.png

small_Logo.png

thumbnail_Logo.png

Logo.png

Platform Integration & Infrastructure Optimization for Pharmaceutical R&D 

Navigating the complexity of integrating advanced C++ features into PyTorch for Windows.

Upgrading C++

Tackling persistent CI challenges to elevate PyTorch's Windows experience. 

Windows CI Test Failures

Ensuring compatibility for devices running on the ARM-based processors under the Linux environment. 

Device Compatibility

Microsoft had a mission to expand PyTorch's reach to the .NET developer community.

.NET Integration Initiative

Transitioned PyTorch to leverage the latest advancements of C++17, supercharging performance and code efficiency.

C++17 Performance Optimization

Addressed multiple CI test failures, ensuring a more stable PyTorch experience for Windows developers.

Enhanced Stability for Windows

Enabled PyTorch's operation on ARM 64 architecture devices like Raspberry Pi and select mobile phones, minimizing cloud reliance.

ARM64 Architectural Harmony

Facilitated PyTorch's reach to the .NET developer community, expanding the user base and fostering wider adoption. 

Integration with .NET via TorchSharp

Ionut

Having shown considerable expertise in maintaining some of the largest open source programs, such as react native and Microsoft PowerToys, Microsoft reached out to us to guarantee developers within the Windows ecosystem could utilize PyTorch to its fullest potential. This project is pivotal in future proofing the platform while fostering a collaborative, open source ecosystem. 

Learn about how Janea Systems Maintains the PyTorch Machine Learning Library on Windows. 

Expert Maintenance of PyTorch for Windows - Enhanced AI Development

large_pytorch-logo.png

medium_pytorch-logo.png

small_pytorch-logo.png

thumbnail_pytorch-logo.png

pytorch-logo.png

ai-machine-learning

open source

Linux-Logo-700x394.png

Python_logo_01.svg.png

NET-Framework-Logo-700x394.png

c#.png

CI CD Icon RED JANEA.webp

Empowering Windows-Based AI Progress

Ensuring core dependencies function correctly on the Windows platform.	

Dependency Stability

Extending Node.js support to ARM64 architecture for Windows.

ARM64 Support

Navigating open-source community dynamics for cohesive development.

Community Collaboration

Building a robust platform for corporate adoption.

Reliability for Enterprises

Addressing speed discrepancies for a seamless experience.

Enhancing Performance

Successfully eliminated frequent bugs and crashes, enhancing Node.js's stability on Windows.

Stability Achieved

Achieved a 40% improvement in file I/O speeds through our optimization efforts.

Performance Breakthrough 

Ensured core dependencies work reliably, supporting a wide range of Windows-specific applications and services.

Dependable Dependencies

Established Node.js as a reliable and preferred solution in corporate environments, broadening its reach.

Enterprise Adoption

Users can rely on native Node.js capabilities like ARM64 compatibility. 

ARM64 Adoption

Stefan

Senior Software Engingeer

Alexis

Founder & Chairman of the Board

When Node.js emerged as a leading server-side JavaScript runtime, it was primarily optimized for Linux and macOS environments, leaving Windows users grappling with stability and performance issues. Realizing the potential of Node.js, Microsoft demanded its high-performance and reliability on Windows to ensure widespread adoption. 
 

Find out why Microsoft had trusted Janea Systems to maintain Node.js on Windows for over a decade.

Maintaining Node.js on Windows Since 2013

medium_nodejs.png

small_nodejs.png

thumbnail_nodejs.png

nodejs.png

Node.js

ISO_C++_Logo.svg.png

Csharp_Logo.png

Ensuring a Smooth Node.js Experience on Windows 

2020_PowerToys_Icon.svg.png

Reviving and updating a historic software utility for modern operating systems.	

Modernizing a Legacy 

Prioritizing, managing, and integrating vast community feedback, contributions and issue reports.	

Community Management 

A variety of libraries and frameworks are used across modules. 

Tech Stack Diversity 

Ensuring seamless performance across Windows devices and releases.

OS Integration

Supported both the x64 and ARM64 CPU architectures. 

x64 & ARM64 CPU Support 

Our team delivered 30 different modules during the cooperation. 

30 Modules Delivered

Effective issue resolution through collective efforts with the broader community. 

25,000 Issues Closed

Strong community engagement, including 7k forks and 5k+ pull requests.

118k stars on GitHub

PowerToys became a live testing ground for Windows native features.

Windows Feature Sandbox

Our team integrated the Semantic Kernel feature to enable generative AI for the Advanced Paste tool.

Generative AI integration

Jaime

Taras

Laszlo

Software Engineer

Seraphima

Anirudha

Janea Systems champions Microsoft PowerToys, enhancing Windows for power users through active development and community engagement. Discover the impact. 

Reviving PowerToys as Open-Source Software

large_2020_PowerToys_Icon.svg.png

medium_2020_PowerToys_Icon.svg.png

small_2020_PowerToys_Icon.svg.png

thumbnail_2020_PowerToys_Icon.svg.png

From Legacy to Windows Feature Sandbox: Reviving PowerToys as Open-Source Software

Overcoming the complexity of the existing testing software.	 

Simplifying Complex Systems 

Reducing test engineers' dependency on software engineers for updates and modifications. 

Breaking Dependency Chains

Streamlining processes to make device testing quicker and more efficient.

Time Efficiency

Enabling non-technical users to manage and adjust testing protocols independently.

Empowering Users

Enabling Cognex to dynamically add and configure new devices for testing, significantly enhancing the flexibility of their testing environment without needing technical intervention for each new device or test scenario.

Enhanced Test Flexibility

By streamlining the testing process and allowing for easier updates, we contributed to a more reliable testing framework. This reliability ensures that industrial vision devices are rigorously vetted under diverse conditions, maintaining Cognex's high standards.

Robust Testing Reliability

The development of a user-friendly interface significantly reduced the learning curve for new users and empowered test engineers to manage testing protocols directly. This empowerment has led to a more engaged and proactive approach to testing among engineers.

Empowerment Through Usability

Nikola

Cognex needed a solution to overcome the complexity and limitations of their existing testing software. Our reputation for delivering custom software solutions in industrial settings led them to partner with us for this project. 

Discover how our software solution revolutionized product testing for Cognex, enhancing efficiency and engineer autonomy in industrial vision systems.



From Kernel to Cloud, with AI-Driven Innovation

What Sets Us Apart

We Make AI Fundamental to Your Business DNA

Projects

4

20+

70+

Latest Blogs

Let's talk about your project

Let's talk about your project