AGROTRADE Group is a vertically integrated agro-industrial Holding with its own production and facilities for grain storage and processing. The main business areas are seed production, crop farming, elevator operations, and trading. The profile crops are corn, sunflower, barley, wheat, soybeans, and rapeseed. Additionally, a range of niche crops is cultivated.

The Group is among the TOP 25 Companies with the largest land bank in Ukraine. It includes 17 agricultural enterprises combined in six agrarian clusters. 

Besides, AGROTRADE owns nine elevators in the Chernihiv, Sumy, Poltava, and Kharkiv regions. They provide 570k tons of simultaneous grain storage capacity.

AGROTRADE Group exports agricultural products to 25 global markets.

The task

The work specifics require the Holding Employees to follow established instructions strictly. Therefore, they often have to refer to the database for answers to work-related questions. At the same time, a large number of regulatory documents, including job descriptions, were being accumulated over a long period and stored without clear systematization in Microsoft 365 SharePoint folders. It took too much time to find the needed document. Sometimes, the file or answer to the work-related question was not found.

To significantly reduce the time for finding necessary information, provide a single access point, and ultimately help Employees avoid a decrease in efficiency due to demotivation caused by searches for documentation, AGROTRADE Group implemented an innovative solution: AI-UniBot Personal Assistant & Corporate Chatbot based on ChatGPT Artificial Intelligence.

The solution

AI-UniBot is an intelligent chatbot that, in addition to fast search, performs many intellectual tasks: it independently analyzes information, creates a meaningful response to the question, and enhances it with links to specific points of the relevant document.

Initially, the Client's Employees were divided in their requirements for the bot's performance. One part expected a quick document search and a short answer to the question. The other part was interested in a quality, detailed response, requiring the bot to conduct a deeper search and analysis of information found in the documents.

Therefore, the Project Team implemented two dialogues in the menu: deep search and quick search. Each dialogue met the Users' requirements and fully satisfied their needs.

How it works

A User who needs a quick search initiates a corresponding dialogue. In a question-answer mode, the User asks questions to the search agent, which is connected to several plugins, including the AgenticSearch plugin. If the answer to the User's question requires referring to the accumulated (indexed) knowledge base, the AI agent uses the AgenticSearch plugin to form and execute search queries. As a result of this search, the AI obtains context, based on which it constructs a response, enhancing it with links to information sources.

If a detailed response is needed, requiring analysis of a large array of documents and possibly comparing versions to determine the relevant one at the time of the search, the User applies the deep search. In this case, the AI agent can independently call the DeepSearch plugin one or more times with different query formulations for the deepest immersion in the topic. For example, if the query involved comparing characteristics of two crops, DeepSearch would be called first for one crop, then for the other. The results will be compared and described in the conclusion prepared by AI for the response.

The DeepSearch plugin, unlike AgenticSearch, can make multiple queries to the knowledge base and independently change the search strategy, as a Human does when searching for information in search engines. During the analysis of the first search page, it can decide to analyze a specific document, obtain the next page, or completely reformulate the search query if the obtained information is irrelevant.

Result

All work-related questions that previously required additional effort to search for documents and time to check them out are now easily and, most importantly, quickly handled by AI-UniBot. For example, it can easily find the instructions and explain how to fill out an application in the 1C system. Or help find the answer to the question of which document accompanies the supply of fuel and lubricants.

The answers provided by the bot are always as relevant, up-to-date, and meaningful as possible, as it is not just a list of links. Analyzing large arrays of documents is done by the bot without Human intervention.

In short, AI-UniBot Personal Assistant & Corporate Chatbot helps every Employee in their work, regardless of their position. Both User needs are taken into account – a fast search for simple questions and a deep iterative search that analyzes files of different formats and provides substantiated conclusions.