The potato was recently featured in WIRED as being a healthful vegetable that doesn’t deserve its bad reputation. KUNGFU.AI recently teamed up with Potatoes USA, a marketing and research organization whose mission is to strengthen the demand of potatoes, to augment their traditional manual approach to correcting inaccurate nutrition information online with cutting-edge AI technology that easily allowed their team to increase reach and productivity.
KUNGFU.AI and Potatoes USA jointly conducted a strategic review of potential methods and platforms on which to launch this initiative. X, on which there are thousands of mentions of the potato each day, emerged as the top candidate, as a thorough manual review of X content was simply too time consuming for their marketing team. KFAI then designed, developed, and deployed an AI-based application to identify and generate responses to critical potato-related narratives that X users were engaged in on a daily basis, relieving the Potatoes USA team of the task of reading through all of the content.
KUNGFU.AI engineers combined multiple machine learning approaches to build this system, addressing the problems of narrative identification, narrative importance, and response generation, the results of which are displayed in a custom application. Tweets are first grouped into narratives based on the similarity of their content, for example, one narrative might consist of tweets that inaccurately claim that potatoes contain bad carbs, while another might concern the cost of potatoes. Once these different narratives have been identified, the system can then determine that the former is important to Potatoes USA, and the latter is not, based on narratives that it has seen before. Surfacing the narratives that are most important to Potatoes USA drastically reduces the volume of tweets that the Potatoes USA team needs to review each day. The user can then easily select a narrative of interest in the application, such as those related to carbs or glycemic index, and leverage generative AI to create a response with the click of a button.
Generative AI technologies such as GPT-4, which powers the Potatoes USA response generation portion of the system, have benefited in accuracy in recent months from the development of Retrieval Augmented Generation (RAG) pipelines. In addition to taking advantage of the vast amounts of data on which the AI models have already been trained, RAG pipelines can also leverage the trove of relevant and factual information about potato nutrition amassed by Potatoes USA to generate candidate responses to narratives for a Potatoes USA user to select or edit. The potato-related knowledge base contains resources such as this article, which outlines research on the nutrient profile of potatoes and qualifies them as a high quality source of carbohydrates. Portions of the article were accessed by the RAG pipeline to inform a model response to a carb-related narrative, refuting the claim that potatoes are ‘a bad carb’.
At KUNGFU.AI, we help businesses harness the potential of AI to augment and streamline their work flows. Yes, the collaboration with Potatoes USA is helping ensure accurate, research-based, information about potatoes is available to consumers, but it’s also a great example of how we work with clients to identify a high impact business problem for AI, develop a proof of concept that demonstrates value, and productionize that solution in an easy to use application.
