AI & Agents
An agent platform where teams build on their own — secure and traceable.
Instead of setting up every agent project from scratch, we build one shared platform: models, tools, data access and approvals as self-service — with tracing, evaluation and monitoring for every request.
Typical situations
- Several teams experiment with agents — each with their own models, keys and tools.
- Nobody can see what an agent actually did, which data it used and what it cost.
- IT is expected to enable GenAI while keeping privacy, permissions and budget under control.
- Good prototypes never reach production because evaluation and monitoring are missing.
- Model and token costs cannot be attributed to a team or use case.
- Every new use case starts from scratch: data access, security and deployment are rebuilt each time.
- Business teams want to build agents themselves but have no safe framework to do so.
- Models, prompts and agent versions are not managed — changes cannot be traced.
What we deliver
- Platform architecture for agents and GenAI — e.g. on Databricks (Mosaic AI, Agent Bricks) or Microsoft Foundry
- Self-service for business teams: templates, approved models, tools and data sources from a catalog
- AI gateway: central model access with permissions, rate limits, guardrails and cost allocation
- GenAI tracing for every request — prompts, tool calls, responses, latency and cost (e.g. MLflow Tracing)
- Evaluation with test sets and LLM judges, production monitoring and user feedback
- Deployment as code with CI/CD across DEV, UAT and PROD
- Catalog of reusable tools and MCP servers for SAP, databases and Microsoft 365
- Operating model: roles, approval process for new agents, training for business teams
Technologies
- Databricks Mosaic AI
- Agent Bricks
- MLflow 3 Tracing & Evaluation
- Mosaic AI Gateway
- Unity Catalog
- Microsoft Foundry
- Azure API Management
- Model Context Protocol (MCP)
- Terraform
Outcome
A platform on which teams ship new agents in days instead of months — fully traceable.