Intuitive GUI replaces time-consuming manual command-line setups
Decentralized data sources for training models without exposing private identities
Advanced project management that allows grouping of unique devices, models, and datasets
Future-ready infrastructure with actively supported frameworks and strict security standards
Summary
A global manufacturer of automotive components worked with Stratpoint to develop a federated learning infrastructure and platform, a groundbreaking orchestration layer built atop the Flower AI framework. In just six months, the team transformed a localized proof of concept into a cloud-hosted, production-grade prototype. By automating the federated learning pipeline, the solution enables the company to train sophisticated AI models across decentralized machines while keeping sensitive data strictly on edge devices.
The project concluded with 100% of requirements met, including a rapid mid-stream pivot to AWS cloud infrastructure and a revamp of the user interface. The company is currently one of the few adopters of federated learning and the new open-source framework, Flower AI.
Challenges
Modern AI often faces a paradox: to get smarter, it needs more data, but the most valuable data is often the most private. Federated learning, a privacy-preserving AI and machine learning technique pioneered by Google, allows algorithms to learn from decentralized data without extracting private information to a central server. The company recognized the immense value of privacy-first AI, but faced significant hurdles in implementation:
- Immature toolsets: The company had selected Flower AI as their framework, but found it lacked the built-in capabilities required for enterprise orchestration and monitoring.
- Manual bottlenecks: Without a centralized platform, every device on which they perform distributed machine learning had to be set up, invoked, and monitored manually, stifling the ability to scale federated machine learning.
- Business risks: Failure to address these challenges would result in the struggle to perform federated machine learning, potentially halting future manageable projects.
- Strategic partnership: The company required a tech partner that could bridge the gap between complex AI research and a functional, user-friendly interface.
The primary user groups of this federated learning project would be AI engineers who deploy custom code, models, and datasets for decentralized training, and R&D engineers and non-technical stakeholders who require a high-level view to monitor resource utilization and system health.
Solutions
Stratpoint deployed an agile team consisting of a tech solutions architect, 2 AI engineers, and 2 full-stack engineers. The team developed a custom infrastructure designed to remove the complexity from federated learning. By building a sophisticated orchestration layer on top of Flower AI, the team provided the company with a flexible, end-to-end system.
Key technical pillars:
- The orchestrator: A centralized hub where users configure projects, upload custom training scripts, and execute jobs remotely.
- Privacy-first Flower power: Utilizing the Flower AI framework in Python to orchestrate model training across multiple machines while keeping data local to the devices.
- Real-time observability: Integration with OpenTelemetry, Tempo, and Grafana for monitoring training progress and resource utilization.
- Flexible deployment: A containerized architecture using Docker and Ansible that allows for deployment on-premises or on AWS EC2 instances.
The Pivot from Local POC to Cloud-Hosted Prototyping
Initially, the project’s goal was a sprint to prove that an open-source federated learning platform could function in a local cluster. After successfully validating this, the vision expanded.
Stratpoint executed a major strategic pivot:
- Cloud migration: The system was rapidly re-engineered to run on AWS, requiring live debugging and a total pivot of the security architecture.
- UI revamp: The project shifted to rapid UI improvements, which included the development of a comprehensive style guide.
- Enhanced security: The team incorporated role-based access, CORS, and password strength requirements to meet near-production standards.
The platform allows users to configure machines to be used for a specific federated learning project for AI model training with decentralized data, deploy model training code on those machines, invoke model training from a centralized platform, and monitor training progress and results.
The company chose Stratpoint for its proven track record in delivering innovative solutions for cutting-edge technologies. By combining deep expertise in AI, backend, and frontend development, Stratpoint moved the company from a struggling initiative to a fully functional, cloud-hosted federated learning infrastructure and learning platform in just six months. For businesses looking to unlock the power of decentralized data without compromising security, the federated learning project serves as the blueprint for scalable, privacy-first AI innovation.
Technolgies Used
