Almost every user’s personal media library these days contains thousands of photos and videos. Manually searching is time-consuming and inconvenient, even with the use of file names and tags. AI is improving, and the demand for smarter solutions is growing – type in “photos from last summer at the beach” and you’ll instantly find what you need.
Software engineer and AI researcher Davyd Maiboroda is working on just such solutions. While leading the machine learning department, he took a fresh look at this problem. He pioneered AI-powered local search, combining neural networks for text and image embedding with vector storage. Now iOS and macOS users can make simple queries in their media library and instantly retrieve the images they need. Maiboroda shares how he created this solution in this article.
Building the Core Architecture
AI-powered image search, selection, or generation is not a new concept, even for users unfamiliar with the latest technologies. However, until recently, solutions that simplified image searches in one’s own media library were lacking. Engineer and AI researcher Davyd Maiboroda took on the challenge of creating such a solution for iOS and macOS. Under his leadership, the team developed a system that locally indexes media files while simultaneously running embedded neural networks directly on the device hardware.
Maiboroda notes that it was important to create a completely local solution, as user privacy is especially important during the rise of AI. To ensure that no data leaves the user’s device, Davyd Maiboroda emphasized this architecture in development without compromising performance.
The resulting system combined two key components: an image converter for generating embedded images/videos and an embedded text model for processing natural language queries. These embedded images were stored in a vector database, enabling fast similarity searching. As a result, users could type or speak a query, such as “birthday party in the garden,” and instantly find matching photos or videos.
Thus, Maiboroda created his own engine, which became a personal search tool for conveniently searching for media content, accessible to any user without any special skills.
Balancing Privacy and Performance
As previously emphasized, Maiboroda and his development team sought a balance between privacy and high performance when creating the local search engine. Even before development began, Davyd prioritized the rule that sensitive image data should not be sent to the cloud until it had been processed and secured. Therefore, with his new engine, the image search process was shifted from indexing to embedding and retrieval directly on the device itself. This gave each user complete control over their media content.
Another aspect, performance, required no less attention from the developers. Maiboroda was aware that using neural networks on iOS and macOS could cause performance issues. The engineer optimized it by lowering its size to 120 MB while preserving accuracy using the MLX accelerator for Apple devices. This struck a balance between efficiency and privacy by enabling the system to react swiftly and fluidly, even to massive media libraries.
From Prototype to Product Integration
Prototyping an AI-powered search engine is one thing, but turning it into a product that works in popular apps is quite another. Davyd Maiboroda brought this idea to fruition, embedding his search engine into real software ecosystems. It became part of productivity tools that users use to manage thousands of files daily. The local search engine is sold as an SDK and has been acquired by several companies. In both cases, it was designed to run in the background, enhancing familiar workflows. End users don’t need to learn anything new to understand how to use this search.
Adapting the engine required attention to cross-platform compatibility. When using MLX on macOS and iOS, the same neural models had to function just as well. Memory constraints, different GPU/ANR optimization techniques, and different user expectations were some of the distinctive features of each environment. To achieve this, Maiboroda optimized the models, aiming to reduce their size and fine-tune their performance to ensure consistently fast and accurate search results across all platforms.
Davyd believes that the more technically complex a feature is, the more important it is to simplify the user experience. Many people are familiar with the concept of AI search precisely because it has become a relatively easy and intuitive process for users. Engineer Maiboroda also wanted to achieve the same goal: maintain a clear and accessible interface. He focused on natural language search functionality. For example, users can type or speak queries like “daughter’s dance competition” or “family dinner in December,” and the search engine will find photos that match exactly. End users won’t see this solution as a groundbreaking, complex technological implementation. Rather, it becomes an add-on to existing software.
MAIboroda showed how cutting-edge AI can go from a technical experiment to a helpful tool that helps millions of end users by effectively bridging the gap between prototype and product.
Redefining Personal Media Discovery
Engineer Davyd Maiboroda’s work demonstrates how AI-powered local search can transform how people interact with personal media. Combining neural search, privacy, and a simple user experience indicates that a local search engine for iOS and macOS is a feasible solution. Furthermore, Davyd managed to preserve the simplicity and naturalness of this complex technology for the end user, which is especially vital now that AI technologies are ubiquitous.
Maiboroda’s engine is a prime example of the new media search standards: it is private enough to guarantee data security, intelligent enough to comprehend context, and user-friendly enough for non-technical people. For the first time, a powerful and customized search is available on consumer devices. This is a step in the direction of AI becoming a necessary but unseen aspect of daily life.
Rashan is a seasoned technology journalist and visionary leader serving as the Editor-in-Chief of DevX.com, a leading online publication focused on software development, programming languages, and emerging technologies. With his deep expertise in the tech industry and her passion for empowering developers, Rashan has transformed DevX.com into a vibrant hub of knowledge and innovation. Reach out to Rashan at [email protected]
