Launch an Agentic RAG Chatbot with MongoDB and Dataworkz

What Is an Agent?

An agent is an artificial computational entity with an awareness of its environment and associated data within the context. These agents can interact in the case of ecommerce or perform a portion of complex tasks as needed.

Retrieval-Augmented Generation

RAG is an approach that augments LLMs with proprietary data so that they can generate more accurate and context-aware responses while also reducing hallucinations.

Figure 1. Conventional RAG

Agentic RAG

Agentic RAG introduces an AI agent-based implementation of RAG. In this model, the agent can access different tools and functions, enabling it to go beyond information retrieval and generation—it allows it to plan. Agents can determine if they need to retrieve specific information or not, decide which tool to use for the retrieval, and formulate queries. These capabilities are crucial as it enables the agent to pull information from multiple data sources, and to handle complex queries that require more than one source to formulate the response. Agentic RAG can enhance customer interactions and streamline processes across industries.

Figure 2. Agentic RAG

For more details see Dataworkz Agentic RAG.

Figure 3. Conventional vs agentic RAG comparison

Impact on the Retail Industry

The retail industry quickly adopted generative AI. In 2024, one quarter of consumers used AI in their shopping experience, representing a 4% increase over the previous year.

Some generative AI use cases across retail include:

• Customer-support chatbots: Gen AI-powered chatbots that provide real-time assistance and are context-aware of business policies, user history, and preferences.

• Personalized product recommendations: Customer recommendations based on their specific likes, needs, and past orders. These personalized recommendations make shopping more enjoyable for customers and increase the chances of a successful purchase.

• Dynamic marketing content: Gen AI applications can generate personalized promotions, emails, and messages for each customer, boosting engagement, sales, and retention.

Introducing Dataworkz

Dataworkz is a RAG-as-a-service platform that transforms how organizations build and deploy AI applications. Its agent-based architecture and graph-optimized retrieval help large enterprises to launch sophisticated RAG applications in hours instead of months.

The platform eliminates the need for specialized AI teams through an intuitive no-code builder that automatically implements best practices in RAG development. Unlike traditional approaches that lock you into early architectural decisions, Dataworkz enables rapid experimentation—you can test different retrieval strategies, prompt variations, and model combinations in a controlled environment before committing to production.

Dataworkz delivers production-ready RAG applications without the traditional overhead of building and maintaining complex AI infrastructure.

This architecture consists of the following key components:

• Graph-optimized knowledge retrieval for complex relationships.

• Agent-based architecture for sophisticated reasoning.

• No-code builder with built-in best practices.

• Full lifecycle support from experimentation to production.

• BYO flexibility—LLM, embedding model, and vector database.

• Enterprise-grade security and scalability.

Leveraging MongoDB with Dataworkz

Dataworkz streamlines the process of creating RAG pipelines by providing a friendly experience to extract unstructured data, configure a chunking strategy, and create vector embeddings. Dataworkz RAG builder also allows developers to choose different retrieval mechanisms—lexical, semantic, or graph, with different thresholds to build the context for answering a user question.

The integration also enables real-time data processing and analytics, ensuring AI models use the most current data for accurate and relevant responses.

Agentic RAG Architecture

To build transformational results, conventional RAG systems, which primarily utilize vector search techniques, must integrate with up-to-date information from operational databases.

MongoDB Atlas Vector Search provides built-in support for vector embeddings, which eliminates the need for a separate vector database, simplifying the architecture and reducing complexity.

With Dataworkz agents, retailers can offer controlled access to MongoDB collections by configuring them as tools. In many cases, customers have an API layer that abstracts the underlying collections. Dataworkz can integrate with REST API or GraphQL.

Additionally, any RAG pipeline configured in Dataworkz can be a tool to an agent. This capability gives agents the ability to understand unstructured data in a SharePoint site, a confluence wiki page, or markdown depending on the user’s question.

Existing retailers with applications that leverage MongoDB as their data platform can benefit from Dataworkz’s close integration with MongoDB and their AI adoption can incorporate agentic RAG into their solutions.

Dataworkz agents can access multiple data sources and use reasoning LLMs to decide which tool to use to answer a user’s question. An agent can access and switch between different MongoDB collections or databases to retrieve structured data like shipping status, customer profiles, preferences, and order histories. Additionally, third party solutions like ERP (Epicor), and CRM (Salesforce) can integrate through APIs exposed via the providers. Together, these tools enable the agent to understand user questions in context and provide personalized, relevant responses.

The following steps explain how a Dataworkz agent works:

1. The Dataworkz agent framework organizes use cases into scenarios, such as answering questions about store policies, searching an order, or providing customer support.

2. The agent uses a reasoning LLM to plan how to answer a question based on the user's query, conversation context, memory, and available tools, such as access to MongoDB collections. The LLM determines a sequence of steps to gather the necessary information. This process is iterative; after each step, the agent checks if it has enough context to provide an answer or if further retrieval and planning are needed.

3. Finally, the agent uses the gathered context to generate a response, take action, or ask the user a clarifying question if needed.

Figure 4. Dataworkz and MongoDB agentic RAG architecture