“Every employee can become a developer”: how Aurora Network is building an AI ecosystem for 17,000 people
The Aurora multimarket chain today has over 1,900 stores across Ukraine and over 17,000 employees. However, behind the large-scale retail, there is a powerful IT transformation. Inside the company, they test Computer Vision on ordinary IP cameras, assemble a «Smart Wikipedia» on self-trained open-source models, and teach non-technical specialists to create useful software using artificial intelligence.
The Aurora multimarket chain today has over 1,900 stores across Ukraine and over 17,000 employees. However, behind the large-scale retail, there is a powerful IT transformation. Inside the company, they test Computer Vision on ordinary IP cameras, assemble a «Smart Wikipedia» on self-trained open-source models, and teach non-technical specialists to create useful software using artificial intelligence.
The R&D director of the network, Oleksandr Voynolovych, told the dev.ua editorial office about why Claude is replacing ChatGPT in Ukrainian retail, how to calculate ROI from AI experiments, and how Aurora creates its own technological product solutions.
AI strategy for 17 thousand people and 4 key areas
The R&D department at Aurora has developed in parallel with the growth of the network itself. Today, the department has over 30 specialists: specialists from the project and process office are engaged in the optimization of large business processes, while other team members work directly on innovations, experiments, and their own developments.
The entire network’s AI strategy is deployed along four main vectors:
AI Adoption: mass employee training that helps them master neural networks and integrate them into daily work.
Velocity: Providing developers with tools to speed up routine code writing.
AI as a product: retraining models to meet the teams' own needs and integrating them into the company’s business processes.
Responsible Use: Legal Policies, GDPR, and Data Security.
«Smart Wikipedia» and Computer Vision
One of the most successful internal projects for the network was «Smart Wikipedia». In large companies, instructions and regulations are usually stored in the form of an array of articles, but at Aurora it was turned into an intelligent knowledge base. Now an employee can ask in simple words in the corporate messenger how to return a product, and the neural network gives a short answer of three sentences instead of pages of text.
They fundamentally did not use open APIs for this, because for 17 thousand people it would cost hundreds of thousands of dollars per month. The team followed the path of digital sovereignty: they took open-source models, created a vector knowledge base, trained the system on their own documents and deployed it on their own servers. This gave a four-digit ROI due to the time savings of employees.
A similar practical approach was applied to computer vision, which the company has been working with for about five years.
«What works in 15 stores is split into 1,500 due to the volume of data and unstable internet. As a result, we came to an effective architecture: the stores have ordinary IP cameras without smart chips, and the entire video stream is processed by local neurons on our servers,» explains Oleksandr Voynolovich.
Thanks to this, the company counts the number of customers with baskets in real time with an accuracy of about 93%, analyzes the conversion from street to check, and records empty shelves. At the same time, personal data is not collected, and people’s faces are not recognized in principle.
Why Claude is replacing ChatGPT and how «vibecoding» is born
The changes also affected the daily back-office tools. Starting with 200 OpenAI Enterprise licenses, the company is now actively migrating to Anthropic Claude. The reason is the higher quality of answers and code, as well as the more convenient five-hour limit intervals instead of the general quota per company, which at OpenAI was quickly exhausted and created bureaucratic delays.
At the same time, the so-called «vibecoding» is gaining momentum in Aurora, when non-technical specialists create software for their own needs.
«We already have real cases where business people create additional services themselves. A person watches a few lessons on YouTube, uses AI prompts to build an application on Localhost, and then comes to us. We check security in two hours and deploy it to internal servers. Automating the old process gives about a 30% increase in efficiency, while complete refactoring and creating a new process using AI speeds up the work many times over. What used to take two months to develop and be rejected due to low ROI is now being built in two days,» notes the R&D director.
Part of the team has to be convinced by their own example, because early adopters in any business are only about 5%. The company’s main message to employees is simple: «AI will not replace you, but a person who knows how to use it will.»
Entering the foreign market: B2B products from Aurora
The main goal of the R&D department is to ensure that innovations are created directly within each business unit, and the central team only supports a safe pipeline for their launch.
At the same time, Aurora is already going beyond internal use and offering its own proven developments to other businesses. Currently, we are talking about four ready-made B2B services:
Traffic Counter: a computer vision-based system for accurate traffic analysis in stores.
Generator Control System: a platform for remote monitoring of generator operation. Convenient when you have a large number of retail outlets in different locations.
Smart Parking: a logistics automation system in warehouse areas that shows the occupancy of warehouse ramps and traffic jams without the need to coordinate drivers over the phone.
LPR & Access Control: an intelligent checkpoint where cameras recognize car license plates and automatically open the barrier for authorized traffic.
Thus, an AI ecosystem is being formed in the Aurora network, in which technologies optimize existing processes and help test new approaches and products. And the R&D department’s own developments become available to other companies.
