Build Federated Learning Pipelines with Ease
Boost Privacy, Minimize Latency and Power Sustainable Applications.
FedMaestro is your backend conductor for Federated Learning (FL) — coordinating secure, decentralized model training across distributed clients with flexibility, observability, and control.
What Is FedMaestro?
FedMaestro is a powerful FL orchestration backend that streamlines the entire lifecycle of federated training jobs. Whether you’re deploying your first model or managing complex multi-round training with hundreds of clients, FedMaestro handles:
Job Creation & Scheduling
Client Coordination
Model Aggregation
Progress Monitoring
Secure Data Flow
How It Works?
1. Job Setup
Admins define job parameters: model, dataset type, aggregation method, and more.
2. Client Registration
Authenticated clients sign up to participate. FedMaestro logs registration history and ensures secure onboarding.
3. Round Execution
Training rounds begin manually or on a schedule. Clients are selected (e.g., random sampling) and the global model is broadcast.
4. Contribution Phase
Clients train locally and upload model updates. Contributions are validated, logged, and queued for aggregation.
5. Aggregation Engine
Client submissions are aggregated using configurable strategies. The new global model is stored and shared.
6. Monitoring & Resilience
Auto-retry if not enough clients respond. Real-time progress summaries and intelligent feedback keep you in control.
Built with a Modern Tech Stack
⚡ FastAPI
Blazing-fast async API server for job orchestration and client interaction.
🔐 Supabase Auth & PostgreSQL
Secure, RLS-enforced storage of jobs, rounds, registrations, and submissions.
🧠 Apache Airflow
DAG-based workflow orchestration for scheduling and retries.
🛠️ Celery + Redis
Background task execution for long-running operations (aggregation, metrics, etc.).
📡 Server-Sent Events (SSE)
Real-time push updates to keep clients in sync.
🗃️ Supabase Buckets
Scalable storage for model weights, logs, metrics, and metadata.
Why choose FedMaestro?
✅ Flexible Architecture
Define your FL pipeline your way — plug in models, strategies, and scheduling logic with ease.
✅ Secure & Auditable
Built-in authentication, row-level security, and structured audit trails for every client and round.
✅ Real-Time Visibility
Stay informed with live updates on round progress, retry events, and aggregated results.
✅ Extensible
Integrate with existing MLOps stacks. Add custom aggregation methods or new client selection policies.
Get Started
Whether you’re building a privacy-first AI system or running cross-institutional training experiments, FedMaestro gives you the tools to orchestrate federated learning at scale.