Harvesting Trust: Methods for Validating Chatbots in Agriculture

In an era where AI is transforming every aspect of agriculture, the creation of trustworthy systems is not just beneficial—it's essential. This talk delves into the rigorous evaluation necessary to build reliable AI tools, specifically focusing on large language models used in agricultural chatbots. Drawing parallels with established QA practices in other industries, we will explore the unique challenges AI presents and outline a robust framework for its assessment. - The Necessity of Scrutiny: Highlighting the importance of Quality Assurance in the iterative improvement of AI, this section will explain why rigorous controls commonly seen in other product categories are crucial for the nascent field of AI, setting the stage for enhanced reliability and user trust. - Unpacking AI Complexities: LLMs introduce specific challenges in evaluation due to their complex, non-deterministic nature. This segment will unpack these intricacies, explaining why traditional QA methodologies fall short and what must be adapted for AI systems. - AI on the Farm: Using Syngenta’s Cropwise AI chatbot for farmers as a case study, this part will detail the critical requirements of delivering relevant, timely, and actionable information to farmers. Examples will demonstrate how AI can assist in decision-making processes from planting to harvest. - Metrics of Improvement: Emphasizing the adage, "you cannot improve what you do not measure," this final section will propose a comprehensive evaluation solution with metrics designed to assess every aspect of the AI workflow repeatedly and under varied conditions to ensure consistency and robustness. Through this session, attendees will gain insights into the specialized needs of QA in AI development, particularly for agricultural applications, empowering them to deploy these technologies with greater confidence and reliability.
Who is this presentation for?
Machine learning and data science practitioners as well as business stakeholders involved in the process of planning and executing on AI projects
Prerequisite knowledge:
None
What you'll learn?
About QA for AI and LLM systems in particular

Yannick Flores is a Senior Data Scientist with over seven years of experience solving complex challenges across industries such as agriculture, R&D, and digital marketing. He is passionate about leveraging data to drive meaningful change, from delivering insightful agricultural information to growers to improving efficiencies in R&D environments. At Syngenta Digital, Yannick leads data science initiatives that scale impactful models globally, including Syngenta’s first commercial large language model (LLM) for seed recommendations and a disease detection solution that helps growers optimize product application and reduce environmental impact. His technical expertise spans deep learning, cloud platforms like AWS, and Machine Learning and Deployment frameworks such as Sagemaker and DVC. Yannick is also dedicated to knowledge-sharing, serving as a data science teacher at Epitech Strasbourg and mentoring aspiring data scientists.