Challenge

Our client, a commercial farming ERP SaaS company, faced a clear limitation in the capabilities of their AI assistant: the model lacked general knowledge about farm operations.

While the system had access to rich quantitative operational data from the ERP database, it was missing the broader contextual understanding required to deliver meaningful insights or act as a true agentic assistant. The AI could answer direct queries, but it lacked the specific expertise in agronomic practices and grounding in operational context needed to provide support in strategic decision-making.

Approach

Out goal was to provide more relevant context to the AI model to improve the reliability, specificity and usefulness of its output. This could be achieved by leveraging a RAG pipeline to equip the AI with structured, contextual knowledge from a vector database.

• We analysed contextual knowledge resources specific to individual farm operations, alongside broader agricultural domain knowledge, to develop a knowledge resource strategy that integrates farm-level practices with general agronomic understanding.
• We built and integrated a RAG pipeline, comprising a vector database, as well as document upload and retrieval logic, to enable the flow of relevant information into the model.
• We designed a user-friendly interface for managing and referencing the ERP's knowledge base that focused on providing visibility into what resources were used as part of the response.
• We rigorously tested and iterated on the system to improve accuracy, grounding, and the quality of responses through user feedback

This approach ensured that the AI could reference not just data, but relevant background knowledge during each interaction.

Result

The enhanced AI agent was now equipped with both operational data and farm-specific contextual knowledge. This allowed it to deliver grounded, actionable, and more accurate insights - transforming its role from reactive Q&A assistant to proactive decision-support tool.

To ground this work with an example… When users queried crop performance in a specific field, the AI could now combine ERP data (e.g. rainfall, fertiliser use) with contextual insights (e.g. field drainage history and agronomic specifications for the crop). This led to richer, more meaningful responses that better supported on-the-ground decision-making.

‍