Multi-Agent AI for Smarter Enterprise Decision-Making
In today's data-driven world, enterprises are looking beyond just dashboards and reports. They are demanding AI systems that can understand the context and refer to multiple data sources and provide actionable insights in real time.
Multi-agent AI system controlled by a supervisor agent built on the Databricks Mosaic AI Framework, powered by Claude Sonnet 4.5, offers exactly that. This architecture enables specialised agents to work collaboratively by governing intelligence to deliver conversational, context-aware insights.
With components like vector stores for semantic search, delta tables for transactional accuracy and governance layers via Unity Catalog, Genie Space through MCP and MLflow, these systems allow enterprises to scale AI responsibly. These components help deliver personalised recommendations and support business users with actionable forecasts.
A perfect demonstration of this approach comes from one of our recent projects with M&S’s Baby Club Division, where we turned traditional analytics into a dynamic, AI-powered experience for both business users and customers.
Author
Nikhil
Mohod
I'm a Data Engineer with 8 years of experience specialising in the Azure data ecosystem. I design and implement scalable data pipelines, lakes and ETL/ELT solutions using tools like ADF, Airflow, Databricks, Synapse and SQL Server. Focused on building high-quality, secure, and optimised cloud data architecture.
Connect with
Nikhil
Mohod
Get a free recap to share with colleagues
What is Lorem Ipsum?
Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an unknown printer took a galley of type and scrambled it to make a type specimen book. It has survived not only five centuries, but also the leap into electronic typesetting, remaining essentially unchanged.
Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an unknown printer took a galley of type and scrambled it to make a type specimen book. It has survived not only five centuries, but also the leap into electronic typesetting, remaining essentially unchanged.
M&S’s Baby Club team wanted more than dashboards. They needed AI that could adapt and personalise, including:
AI-driven insights from sales and product data.
Supporting business users with advanced analytics and forecasting.
Delivering personalised recommendations for customers based on Blue (baby boy) or Pink (baby girl) products.
Enhancing guest engagement through intelligent product discovery.
Ensuring all workflows remained secure and governed.
Multi-Agent AI in Action
Cloudaeon implemented a multi-agent architecture orchestrated by Claude Sonnet 4.5. By leveraging Databricks and MCP Server, each specialised agent contributed a specific capability:
Governing agent (Claude Sonnet 4.5)
Central orchestration across agents.
Choose between querying vector stores, delta tables or both.
Ensured Baby Club context (Blue/Pink) is applied correctly.
Prompt optimisation agent
Refined user queries into precise, actionable requests. For Example: “What’s trending for baby boys?”, “Show top-selling products in the Blue category.”
Embedding & retrieval agents
Converted queries into semantic embeddings stored in the Databricks vector store index.
Executed hybrid retrieval from semantic (vector) and structured (Delta) data.
Delivered factually accurate and contextually relevant insights.
Governance and evaluation
MLflow evaluations tracked response relevance and performance.
AI gateway secured model/API access.
MCP (Model Context Protocol) ensured smooth integration between agents and backend tools.
Together, this framework enabled real-time and AI-driven insights that transformed how business users and M&S customers could interact with the Baby Club offering.
How Multi-Agent AI Works: Turning Queries into Intelligent Conversations
User query enters via AI Gateway
Every interaction, whether from a business user or customer, is securely processed through the AI Gateway. This manages the access and governance.
Governing agent (Claude Sonnet 4.5) identifies the user profile
The system determines who’s asking and applies the context, whether it is a business user seeking analytics or a customer reviewing orders or a guest browsing products.
Governing agent chooses the retrieval strategy, vector, delta, or hybrid, depending on the query. Claude Sonnet 4.5 decides whether to draw from:
Vector stores for semantic and AI-driven recommendations.
Delta tables for factual and transactional data.
Or a hybrid approach combining both for richer insights.
Specialised agents execute the workflow: Prompt optimisation - embedding -retrieval/search
The prompt optimisation agent refines the query, while the embedding agent transforms it into semantic form and the retrieval & search agent fetches relevant data from both structured and unstructured sources.
Governing agent compiles a Baby Club-specific answer
Claude Sonnet 4.5 merges the retrieved data and applies Blue/Pink product logic, ensuring the response is accurate, contextual and aligned with the Baby Club categories.
MLflow evaluations track quality
Each response is measured for relevance, accuracy and performance, allowing the system to continuously improve through feedback loops.
Response delivered as a natural AI conversation
Finally, the answer is presented conversationally, transforming traditional queries into intelligent, human-like interactions that empower business users and engage customers.
Conclusion
This project highlights how multi-agent AI systems built on Databricks Mosaic AI Framework can revolutionise enterprise decision-making.
By combining structured (Delta) and unstructured (Vector) data, orchestrating specialised agents with Claude Sonnet 4.5 and embedding governance through Unity Catalog, MLflow and AI Gateway, we enabled:
AI-driven growth and insights replacing static analytics.
Personalised experiences for customers and contextual decision support for business users.
Scalable and modular architecture ready for enterprise-wide adoption.
The Baby Club use case demonstrates that multi-agent AI is not just a technical innovation, it’s a blueprint for modern enterprises seeking intelligent, responsible and actionable AI across domains like retail, finance, healthcare, etc.
With Cloudaeon’s expertise in AI orchestration, governance and enterprise-scale deployment, organisations can now move beyond traditional analytics and embrace AI systems that think, converse and create tangible business value.
