RAG-Powered Enterprise Search Across A Database

Problem

Our client, a multinational distributor, relied heavily on Microsoft Dynamics365 ERP to manage inventory, suppliers, and financial operations. Although the system contained all essential business data, employees faced constant delays when trying to access it.

Frequent lookups such as stock levels, order statuses, or invoice tracking required navigating through multiple ERP modules. Reference materials — including product specifications, warranty policies, and compliance guidelines — were scattered across SharePoint and internal knowledge bases.

Traditional keyword search returned inconsistent results, and constant querying of the ERP risked slowing down system performance.

The company approached us to develop a secure conversational assistant that could deliver fast, natural-language answers while maintaining real-time accuracy and protecting the ERP’s performance.

Solution

We developed an AI chatbot powered by Retrieval-Augmented Generation (RAG), designed specifically for Dynamics365 ERP. The solution uses a hybrid data integration model, balancing real-time ERP connectivity with semantic search over static information.

Direct ERP Integration (Real-Time Data)

The chatbot connects directly to Dynamics365 APIs to retrieve live operational data such as purchase order statuses, customer invoices, and current stock levels.

This ensures that time-sensitive information is always accurate and instantly accessible to employees.

Intermediary Database (Static and Semi-Static Data)

For less frequently updated data — such as product descriptions, warranty details, and compliance documents — we built an intermediary storage layer.

Information is extracted from both Dynamics365 and SharePoint and stored in a PostgreSQL database with pgvector embeddings for semantic search.

This approach allows employees to ask contextual questions without generating unnecessary load on the ERP system.

Hybrid Query Orchestration

The chatbot uses intelligent orchestration logic to determine which source to query based on the employee’s request.

• Operational queries are routed to the ERP APIs.
• Reference or contextual queries are handled by the vector database. When needed, both sources are combined into a single, concise, and easy-to-read response.

Conversational AI

The system is powered by GPT-4o, deployed through a private Azure OpenAI endpoint for data security and compliance.

It interprets complex natural-language queries from employees and returns structured, contextually relevant answers.

User Interface

The chatbot is integrated directly into Microsoft Teams, allowing employees to access ERP data without switching applications.

 Responses include inline references to ERP records or SharePoint documents, helping users verify the source of each answer.

Technology Stack

• ERP Integration: Microsoft Dynamics365 APIs
• Database & Search: PostgreSQL + pgvector (semantic + keyword hybrid search)
• LLM Engine: GPT-4o via Azure OpenAI private endpoint
• Interface: Microsoft Teams chatbot with streaming responses
• Security: Azure AD Single Sign-On and role-based access control

Results

 The chatbot transformed ERP interaction into a fast, conversational experience:

• Real-time accuracy: Employees can access live ERP data instantly without manual navigation.
• 40% reduction in time spent searching for supplier, order, and invoice information.
• Improved system performance: The hybrid model prevents excessive ERP API calls.
• Rapid adoption: Within two months, over 65% of procurement and finance employees were actively using the chatbot.

The result is a secure, enterprise-grade conversational assistant that turns Microsoft Dynamics365 ERP into a natural-language interface — delivering both real-time operational data and contextual company knowledge as easily as asking a colleague, but with the accuracy and speed of AI.