This white paper presents a successful implementation of a Retrieval-Augmented Generation (RAG) application, developed by Computomic in collaboration with a leading Manufacturing enterprise. The project demonstrates how the Databricks Data Intelligence Platform was used to transform unstructured product knowledge into a real-time, AI-powered assistant integrated within Microsoft Teams. The goal was to solve a critical sales enablement issue, enabling frontline employees to access accurate product information across thousands of permutations and formats instantly.
Introduction
In today's fast-paced enterprise landscape, having real-time access to accurate and contextual product information is essential for business success. Our client faced challenges in product recommendation due to the fragmented nature of their product documentation scattered across hundreds of manuals, specification sheets, and tribal knowledge. To address this, Computomic built an enterprise ready GenAI solution powered by Databricks, Azure, and Microsoft Teams, leveraging Retrieval Augmented Generation (RAG) architecture to enable a contextual, searchable interface for complex product data.
To address this, Computomic built an enterprise ready GenAI solution powered by Databricks, Azure,
and Microsoft Teams, leveraging Retrieval Augmented Generation (RAG) architecture to enable a contextual, searchable interface for complex product data.
Business Challenges
The customer sought to:
- Evaluate Databricks as the core data platform for their LLM and AI readiness
- Handle heterogeneous data sources like SAP, Salesforce, IoT, and document
- Identify high impact use cases to validate their commitment to Databricks as their enterprise-wide Data & AI platform
The selected use case focused on improving sales
productivity:
- The client has 160,000+ product SKUs with thousands of configuration permutations, making it hard to search and find the right product
- Salespeople were unable to confidently recommend the right products
- No centralized repository: knowledge was scattered and inaccessible
- Sales managers were manually fielding queries, slowing response times
Solution Overview
Computomic designed and implemented an AI-powered Teams chatbot with RAG capabilities. The solution integrated Databricks with Azure services to create a secure, scalable, and intelligent assistant capable of:
- Summarizing documentation
- Performing vector similarity search
- Generating contextual answers from structured and unstructured data
Architecture & Technical Components
High-Level Architecture
Data Sources → Processing → RAG System → Genie Space → Teams Bot
Data Sources:
- Structured: Product tables
- Unstructured: PDF manuals, specification documents
Key Technologies:
- Databricks: RAG pipeline, Genie Space, vector search
- Azure Bot Service: Teams integration
- Azure Web App: Bot hosting
- OAuth: Secure authentication & RBAC
- Custom Serving Endpoints: Real-time API response handling
The Technical Challenge
Our enterprise client had 10,000+ PDFs and structured SAP HANA data, but their existing search could only return products by name and
specifications. Sales teams needed contextual intelligence: "What product that we have will be well suited for a place like Florida (Humid Conditions)? instead of generic product lists.
The Databricks-Powered Solution
Core Architecture Components
Databricks GenAI Serving Endpoint
- Custom RAG pipeline deployed as a managed serving endpoint
- Real-time inference with sub-30-second response times
- Automatic scaling based on query volume
Databricks App Connector Integration
- OAuth2 authentication with user-level access control
- Seamless token management for enterprise security
- Fine-grained permissions tied to organizational roles using Unity Catalog
Hybrid Data Processing
- Structured Path: SAP HANA,SQL Server →Databricks Delta Table → Databricks Genie Space
- Unstructured Path: PDF ingestion → Databricks Volume → Vector embeddings → Vector Search
- Real-time data fusion for comprehensive recommendations
If you want the complete document, ¡download your copy now!