Case study: Hypermode + AgroPatterns

In industrial-scale agricultural operations, growers are beset by challenges to produce high-quality yields during peak seasons. AgroPatterns leverages Hypermode's knowledge graph engine and tools to help agronomists effectively manage the health of their crops and forecast the maturity of their harvests.

About the company

AgroPatterns offers an integrated Monitoring & Evaluation suite of products that leverage data collected in the field through visual inspections as well as collected automatically by IoT devices, in order to provide relevant and timely insights to industrial horticulture operations (i.e., fresh flowers). To meet the demands of the fresh flower industry, growers meticulously schedule and manage plant growth cycles to meet peak-driven demand, ensuring they can deliver the right quantities at the highest quality. AgroPatterns is a data capture and decision-support system designed to meet the needs of industrial ornamental flower farms. This system is critical for ensuring compliance with stringent pest control standards, enabling over 90% of flowers to meet export requirements for markets in the U.S., Europe and Asia.

The challenge

Industrial farms strive to maximize productivity by gaining deeper insights into their crops, empowering them to implement targeted strategies to mitigate damage, such as applying fungicides or pesticides. However, excessive use of these measures incurs both economic and environmental costs. To minimize reliance on pesticides, growers must closely monitor growing conditions, while agronomists continuously adjust care plans to sustain yields and promote sustainable farming practices.

The biggest challenge facing agronomists today is the ability to efficiently capture data and reliably receive intelligence related to the captured data. Scouts inspect plants and plant beds for issues on the growing room floor. They take notes on an offline mobile application, syncing the data once they get back to the office. The Agropattern IoT devices in each greenhouse take real-time temperature and humidity readings. Agronomists then use this data to adjust the care plan for each bed in each greenhouse to reach the crop's growth goals.

“What I need is the ability for users of AgroPatterns to communicate with the system in their own language,” said Gabriel Coch, Founder and CEO of AgroPatterns. “In order to be effective, the technology needs to use tools and language that are familiar to people that work in the industry.”

Constantly changing environmental factors can negatively impact output or even spoil an entire crop. Agronomists must be able to quickly and optimally manage temperature conditions and humidity. Therefore low-latency data collection and real-time, contextual responses were critical to incorporate into the AgroPatterns app.

The solution

Coch quickly saw the potential of AI as a means to provide an ideal way of ingesting, analyzing and leveraging large amounts of data collected in the field by his clients. While he initially explored ChatGPT and other similar tools, he quickly realized that his use case required a much more specific language model.

While searching for solutions, Coch found in Hypermode both a strong partner and the right combination of technologies that made sense for his use case. Coch's experience in route optimization algorithms quickly helped him realize that Hypermode's graph database, Dgraph, and code-first framework would enable the real-time horizontal scalability and flexibility required for his needs.

The collaboration between AgroPatterns and Hypermode resulted in an innovative approach that utilized data, small models, vector search and LLMs. A key component was also the creation of a customized language model by AgroPatterns, designed to accurately interpret agricultural terminology.

The first step involved adding a natural language query capability into the existing AgroPatterns application. This new user interface allowed farmers and scouts to ask questions about general agronomic practices or specific field conditions they encountered. Hypermode quickly iterated by exposing an API and encapsulating the AI logic within functions. Responses to user queries are constructed from AgroPatterns knowledge, which is extracted from text files and loaded into Hypermode's Dgraph database, creating a specialized knowledge graph. This enables the solution to discern, for example, that "Acrobat" refers to a fungicide in an agricultural context, rather than a PDF reader.

Within weeks, AgroPatterns integrated a GraphRAG feature into their existing application. This feature stores questions and answers within a graph structure, utilizing small, cost-effective models to compute vector embeddings. Now, the application leverages Dgraph semantic search based on vector indexing to identify similar questions in the Q&A knowledge base before generating new responses from the documentation corpus. Additionally, a fallback mechanism uses a large language model to answer user questions, incorporating a review queue managed by AgroPattern experts to ensure that the system is robust and continuously improving. To further enhance accessibility, AgroPatterns developed a WhatsApp integration utilizing the same API.

The AgroPatterns team were also able to create a comprehensive knowledge graph that integrates product usage documentation, agricultural expertise, and historical farm data. The system now tracks multiple indicators per bed, including mites, mildew, aphids and other crop-specific pests.

The results

AgroPatterns enables growers to make data-driven decisions, maximize their yield, and avoid costly or toxic countermeasure use. By replacing the use of paper and pencils with electronic notes synced to the cloud, the AgroPatterns app:

• Empowers agronomists to work with real-time data rather than wait for tabulated records, thereby reducing the need for insect or fungal countermeasures, as well as cutting costs and avoiding worker exposure to hazards.
• Gives users access to a system optimized for the unique needs of the fresh flower industry, reducing waste and increasing profit.

As early adopters of the Hypermode platform, the AgroPatterns team is actively working to enhance their app by continuing to transition towards natural language processing for all user interactions and planning voice-based data collection to replace manual entry. In partnership with Hypermode, they are building a recommendation system, specifically for the usage of pesticides that have an impact on operational costs.

The AgroPatterns team is also exploring sensor integration for real-time temperature and condition monitoring. A key focus is making complex data accessible through a conversational interface, with the goal of enabling users to extract valuable insights from daily collected data.

The end goal for Coch and AgroPatterns is to overcome AI adoption hesitancy among agricultural workers.

“Hypermode is making this possible by providing the underlying technology and making it easier to integrate natural language features into AgroPatterns,” said Coch. “This is part of the revolution in how farming will be conducted, and I'm excited to continue this work with the Hypermode team.”