Complex platforms face a universal challenge: as they grow more powerful, they also become harder to use. Features accumulate across modules, and navigation deepens, causing users to spend more time searching than working.
We spoke with Solomiia Bilyk, our Data Scientist, about how a multi-agent AI approach can make complex platforms easier to use and how this approach is reshaping enterprise software.
With 6+ years of experience delivering AI-driven solutions in retail, productivity, document analysis, and edge computing, Solomiia specialises in both data science and engineering. Her work focuses on building practical tools and frameworks for computer vision, NLP, and optimisation.
What challenges emerge when platforms grow and evolve?
Solomiia Bilyk: Modern enterprise platforms typically grow in three dimensions:
- Depth: Individual features become more sophisticated, with advanced capabilities that require specialised knowledge to use effectively.
- Breadth: New modules are added to serve different user groups, each with its own interface, workflows, and data structures.
- Speed: Development velocity means features evolve faster than documentation, creating knowledge gaps even for experienced users.
The result is a system that's incredibly powerful but increasingly difficult to navigate. Users ask the same questions repeatedly:
- "Where do I find this?"
- "How do I create that?"
- "What insights can I get from all this data?"
Traditional solutions — better menus, improved search, comprehensive documentation — provide incremental improvements but don't solve the fundamental problem: users think in terms of outcomes, while data platforms are organised by technical architecture.
How do you bridge the gap between platform complexity and user needs?
SB: Instead of forcing users to learn the platform's structure, we built a conversational interface that handles three critical use cases:
- Product discovery for newcomers exploring the system. Users ask how features work and get answers grounded in actual documentation rather than hallucinated information.
- Workflow simplification for familiar users who want faster task completion. They can create items, find specific data, and navigate complex hierarchies through simple commands instead of multi-step processes.
- Insight generation for advanced users managing data across multiple modules. The system combines information from different parts of the platform to deliver comprehensive analysis.
A single conversational interface serves all three needs, meeting users where they are in their journey with the platform.
For example, at ELEKS, we recently solved this problem for a client whose rapidly evolving system had become a maze of specialised tools, each designed for different user groups. The solution wasn't simplifying the platform — it was creating a single intelligent entry point that makes complexity invisible.
How is the system structured to handle different types of user requests?
SB: The system uses five specialised agents, each handling a distinct type of request:
- Intent router: The first layer classifies what users are trying to accomplish — searching for items, creating objects, understanding features, comparing data, or extracting insights. Precise classification enables targeted processing with domain-specific tools and optimised prompts.
- Finder agent: This component solves the "I remember the concept, not the location" problem using semantic search that matches meaning rather than exact keywords. It filters by metadata, including dates, tags, and user associations, then scores results based on relevance and business rules. This bridges the gap between fuzzy human memory and rigid database structures.
- Action agent: When users request item creation, modifications, or system actions, this agent executes validated operations through strictly typed API calls. No guessing, no hallucinating features that don't exist, just safe, predictable operations that either succeed correctly or fail with clear error messages.
- Knowledge agent: For questions about platform capabilities, this agent uses retrieval augmented generation (RAG) to pull answers from actual help documentation. When users ask, "How do I export a report?" or "What can the analytics dashboard do?", they get grounded, accurate responses referenced from official guides.
- Insight agent: For analytical requests, this component compares items across dimensions, identifies patterns in user data, and generates summaries that transform raw information into actionable conclusions. It handles requests that require synthesising data from multiple platform modules.
What are your recommendations for deploying AI agents in production?
SB: Here's what we learned from building and scaling our multi-agent system.
1. Start with user outcomes, not system structure. The biggest value shift: users stopped needing to understand where information lives or how workflows are structured. They describe what they want to accomplish, and the system handles navigation and execution.
To maximise AI agent value, investigate the main pain points of different user groups — newcomers discovering the system, regular users seeking efficiency, and advanced users needing comprehensive insights. This ensures the solution meets actual needs rather than demonstrating technical capabilities.
2. Balance model capability against cost. During development when requirements changed weekly, we used smaller, cost-efficient models. Fast iteration with lower expenses made more sense than perfect accuracy on rapidly evolving prompts.
As the system stabilised and usage patterns became clear, we upgraded specific agents to more powerful models where the return on investment justified higher costs. Match model capability to task requirements and usage volume rather than defaulting to the most expensive option.
3. Prioritise predictability over creativity. For agents that take action, reliability is the only metric that matters. Deterministic, accurate behaviour is far more valuable than eloquent or creative responses.
Enforce this through strict output schemas, rigorous validation, and a policy of failing safely, saying "I can't do that" rather than guessing. Users need to trust that the same request produces consistent results and that the system won't execute incorrect operations.
4. Build framework-independent architecture. Artificial intelligence frameworks are incredibly useful for development speed, but they evolve rapidly with frequent breaking changes and shifting best practices.
Build thin abstraction layers for routing, retrieval, execution, and knowledge access that don't depend on framework internals. Treat frameworks as interchangeable plugins rather than architectural foundations. This keeps your system maintainable and portable as tools and standards evolve.
What is your vision for the future of complex enterprise software?
SB: Multi-agent AI systems solve platform complexity not by simplifying the underlying system, but by creating an intelligent interface layer that translates user intent into system actions, whether that's navigating features, executing workflows, or managing data engineering tasks.
The platforms that will thrive aren't necessarily the simplest. They're the ones that make complexity manageable by meeting users at the level of intent rather than requiring them to understand system architecture.
When users forget they're talking to AI and simply focus on getting work done more easily, that's when you know the complexity problem is truly solved. Technology disappears, and what remains is effortless productivity in software development.
Related insights
The breadth of knowledge and understanding that ELEKS has within its walls allows us to leverage that expertise to make superior deliverables for our customers. When you work with ELEKS, you are working with the top 1% of the aptitude and engineering excellence of the whole country.
Right from the start, we really liked ELEKS’ commitment and engagement. They came to us with their best people to try to understand our context, our business idea, and developed the first prototype with us. They were very professional and very customer oriented. I think, without ELEKS it probably would not have been possible to have such a successful product in such a short period of time.
ELEKS has been involved in the development of a number of our consumer-facing websites and mobile applications that allow our customers to easily track their shipments, get the information they need as well as stay in touch with us. We’ve appreciated the level of ELEKS’ expertise, responsiveness and attention to details.
