A New Era of Enterprise AI: Project Fleming Now Integrated with Mosaic AI by Databricks

Project Fleming—the open-source AI discovery framework can now be enahcnes with Mosaic AI by Databricks. Empowering teams with open-source flexibility and enterprise-grade reliability

With this integration, users can now harness the exploratory power of Project Fleming—known for its intuitive discovery tools, branching logic, and open-source flexibility—directly within the Databricks Data Intelligence Platform.

🔍 Key Features of Mosaic AI available in Project Fleming

🔐 Permission and Rate Limiting

Purpose: Controls who can access model endpoints and how much they can use them.

Functionality: Allows administrators to define access policies and rate limits per user or group, ensuring fair usage and preventing abuse.

Support: Available for most endpoint types, including external models, foundation model APIs (both provisioned and pay-per-token), and custom models. Not supported for Mosaic AI agents.

📦 Payload Logging

Purpose: Enables monitoring and auditing of data sent to model APIs by logging requests and responses into inference tables.

Functionality: Captures and stores input/output payloads for each model call, enabling traceability, debugging, and compliance auditing.

Support: Fully supported across all endpoint types, including Mosaic AI agents.

📊 Usage Tracking

Purpose: Tracks operational usage and costs of model endpoints using system tables.

Functionality: Aggregates and reports on endpoint activity, including call frequency, latency, and cost metrics, to support budgeting and optimization.

Support: Supported for all endpoint types except Mosaic AI agents.

⚖️ Traffic Splitting

Purpose: Load balances traffic across multiple models.

Functionality: Useful for A/B testing or gradual rollouts.

Support: Supported for external, provisioned foundation model APIs, and custom model endpoints 1.

🔍 Further Features of Mosaic AI available with Databricks

🔐 AI Guardrails

Purpose: Prevents unsafe or non-compliant data from being processed.

Functionality: Filters out harmful or unwanted content in both requests and responses.

Support: Available for most model endpoints except Mosaic AI agents 1.

🔁 Fallbacks

Purpose: Ensures reliability by minimizing production outages.

Functionality: Automatically switches to backup models or endpoints if the primary fails.

Support: Currently only supported for external model endpoints 1.

🌐 Why This Matters

This integration isn’t just a technical upgrade—it’s a strategic enabler. Enterprises can now: - Build compound AI systems faster - Ensure governance and compliance from day one - Empower teams with open-source flexibility and enterprise-grade reliability

🤝 Get Involved: Contribute to Project Fleming

Project Fleming thrives on the power of community. Whether you're a developer, researcher, data scientist, or AI enthusiast, your contributions can help shape the future of open-source AI discovery.

Here’s how you can get involved: - ⭐ Star the project on GitHub - 🧠 Join discussions and share your ideas in the community forums - 📢 Spread the word by writing blog posts, tutorials, or hosting meetups

Together, we can build a smarter, more open AI future.