Building AI That Works: How Lexis Solutions Delivers Practical Integration and Trusted Agentic Systems

In the rush to adopt artificial intelligence, many companies find themselves sitting on mountains of data yet still waiting for real results. Tools pile up, projects drag on, and the gap between promise and practical value only widens. What stands out are the teams that cut through the noise, choose one high-impact process, and turn it into a working system people actually use.

Lexis Solutions has built its reputation on exactly that approach. This boutique development firm does not simply follow a client’s list of requests. Its experienced team steps in with clearer thinking, suggesting smarter paths that save time, money, and the costly rework that so often follows. Working across North America and Europe with both startups and established enterprises, Lexis delivers focused AI tools on a fixed scope and fixed fee, typically within fourteen days.

As a Select partner in the Claude Partner Network, the firm brings Anthropic-certified practitioners and partner-channel support to every engagement. Each project begins with a discovery session that identifies the single highest-value process. The team then connects one existing data source, builds one production-ready AI automation, and hands over a lightweight interface complete with usage tracking and team training. The result is never a prototype. It is a complete, working system that scores call scripts, flags invoice overcharges, generates capability statements, or answers queries across large knowledge bases, ready for daily use from day one.

Recognized by CIO Bulletin among the “Top AI Integration and AI Agent Development Companies 2027,” Lexis Solutions continues to stand out for clear communication, fair rates, and the quiet confidence of a team that builds what works. CIO Bulletin had the distinct privilege of interviewing Bilyal Mestanov, Co-Founder and CTO of Lexis Solutions, who shared valuable insights into the principles and practices that define the company.

Interview Highlights

As artificial intelligence moves from experimentation to real-world business transformation, what inspired Lexis Solutions to embrace this shift, and how is your vision shaping the next chapter of the company?

Lexis Solutions started with AI before it became a services company. In 2016, four of us met as students at HackFMI, the Sofia University hackathon, and built a machine learning product that analyzed emotion in Twitter data. We called it Lexis. It won Best Startup at Junior Achievement's Rising Stars competition, but we quickly learned that local businesses were not yet ready to hand their data to an AI system. We started building software for clients instead, and that became Lexis Solutions.

That early lesson still shapes how we work. The hard part of AI is rarely the model. It is the data, the permissions and the trust around it. Today Lexis Solutions is an AI integration and AI agent development company, a Select partner in Anthropic's Claude Partner Network, and around 90% of our engineers are Claude Certified Architects. Our next chapter is focused on bringing AI into established companies that have years of data and processes behind them, in a way their security, legal and operations teams can sign off on.

Your philosophy, “Anyone can follow directions. We make them better,” challenges the conventional software development model. How does this mindset help you uncover smarter solutions and deliver greater value to clients?

Most software vendors are paid to execute a specification. The problem is that a specification written before anyone has looked closely at the data is often wrong, and with AI projects it is wrong more often than not. Our engineers are expected to say so. If a client asks for a chatbot and the real bottleneck is that their reports sit in a data warehouse that only two analysts can query, we will point that out and propose the smaller, more useful system.

This saves clients’ money because the expensive mistakes in AI projects happen at scoping, not in the code. It is also why our client relationships last: Lexis Solutions holds a 5.0 rating on Clutch, and clients such as OnnoPlus have worked with us for years, growing from one system to a whole platform.

From AI-native applications and multimodal platforms to RAG systems and multi-agent orchestration, Lexis Solutions operates across several areas of AI innovation. Which capabilities best define your expertise and what sets your approach apart in a rapidly evolving market?

Our core expertise is AI that works on a company's own data, under the company's own rules. That covers three areas: retrieval-augmented generation (RAG) knowledge systems that cite their sources, governed AI assistants that sit on top of business systems, and agentic workflows that take actions across a client's stack.

Two projects show the range. For Finansi.bg, Lexis Solutions built a financial data platform that processes more than 2 million company reports using Anthropic's Claude models, Voyage AI embeddings and a ChromaDB retrieval layer. It analyzes 416,000 companies every month, and fewer than 8,000 of the 2 million documents needed a human to step in. For Barton Malow, a US construction company, we built a governed, multi-model AI assistant that answers business questions by running SQL against the company's warehouse through the Model Context Protocol (MCP), instead of relying on the model's memory.

What sets us apart is that we are also a data engineering and web scraping company. The pipelines that feed our AI systems are usually ones we built and maintain ourselves, so we are accountable for the whole chain, not only the model layer.

How do you help organizations move beyond experimentation and embed AI into existing systems, products, and workflows to achieve measurable results?

We start with one workflow, one data source and one success metric, and we build it into the tools people already use rather than into a separate AI portal. Adoption comes from fitting into Microsoft Teams, the CRM or the reporting tool a team opens every morning.

At Barton Malow, routine questions used to go into a queue for a BI analyst. The assistant we built was rolled out across the organization rather than to a small pilot group, and in testing it answered 62% of real business questions better than native natural-language-to-SQL tools. In one benchmark question about open projects, our pipeline returned all 21,314 matching records while the direct approach returned 370, because it had picked the wrong table.

For Onno Plus, a Berlin-based online reputation company, we built an AI-first CRM that became the operating system of the business. Sales grew by 210% and order booking became 99.8% faster. Those are the numbers we ask clients to hold us to: time saved, revenue, error rates, not demo quality.

How is Lexis Solutions helping businesses harness this next generation of automation to solve complex operational challenges?

The next generation of automation is agents that read, decide and act across several systems, with people approving the steps that carry real consequences. At Onno Plus, the system scrapes new reviews from Google, Trustpilot and Kununu every day, analyzes them, and drafts legal removal complaints based on historical cases. A lawyer reviews every complaint before anything is sent. That combination of automation and a hard human checkpoint is what lets a legal team handle far more cases without lowering quality.

We are applying the same pattern in construction and preconstruction, with tools such as cost estimators and fleet invoice overcharge detection, and in large-scale speech-to-text and text processing, where hours of recorded calls and documents become structured data a business can search and act on.

As organizations explore autonomous AI, how do you balance innovation with reliability, data security, human oversight, and the need for AI systems to deliver consistent, trustworthy outcomes?

We treat AI governance as an engineering problem, not a policy document. The rules that matter are enforced in the system itself.

In the Barton Malow assistant, each user queries the data warehouse under their own identity, and permissions are enforced by the warehouse. If someone has no access to finance data, the query never runs, regardless of how the question is phrased. Arithmetic goes to a deterministic calculator rather than the model. Every answer is produced by executing a tool against real data, so it can be traced.

We also design for the cases where AI should not decide alone. At Finansi.bg, the pipeline flags uncertain pages for manual review. At Onno Plus, a lawyer approves every legal document. Add evaluation loops after launch, and clients get systems that behave consistently and that their compliance teams can actually audit. AI governance is now one of the fastest-growing parts of our work at Lexis Solutions.

Your 14-day AI development model promises a production-ready workflow with a clearly defined scope and fee. What makes this approach effective, and how can it help businesses validate AI’s value before investing in larger transformation initiatives?

The 14-day model at Lexis Solutions delivers one AI-powered workflow in production in 14 days, at a fixed fee between EUR 2,000 and EUR 5,000 agreed before work starts. The client gets a lightweight interface, one connected data source such as a CRM, database or spreadsheet, one AI automation, user access for their team, usage tracking and training. It is not a prototype.

It works because it forces the decision that most AI initiatives avoid: which single process is worth automating first, and how success will be measured. Typical builds include call script adherence scoring for support teams, preconstruction cost estimators and knowledge base assistants. After 14 days the client has real usage data and can decide what to scale with evidence, not a slide deck. The fixed scope and fixed fee also make it easy for a CFO to approve.

As Lexis Solutions continues its growth journey, what enduring principle will guide your next phase of innovation, and what lasting impact do you hope to create for businesses through intelligent software and AI-driven solutions?

Build what works best, not only what was asked for. That principle has guided Lexis Solutions since we were four students with a machine learning prototype, and it matters more as AI tools multiply. Part of it is being honest when AI is the wrong answer to a problem.

The impact we want is simple to describe. Many established companies are sitting on ten or twenty years of documents, reports and operational data that almost nobody can use. We want every employee in those companies to be able to ask a question and get a correct, sourced answer, limited to what their role is allowed to see.

Looking toward 2027, which emerging developments in AI integration and agentic systems do you believe will have the greatest impact on businesses, and how should organizations prepare to adopt them strategically?

Three developments will matter most in 2027. First, agents connected to business systems through standards such as the Model Context Protocol, which make it practical to give AI controlled access to tools and data. Second, AI governance moving from optional to required, as EU AI Act obligations phase in and boards ask who is accountable for automated decisions. Third, unstructured data such as calls, meetings and scanned documents becoming usable at scale through speech-to-text and document AI.

To prepare, companies should map their data and access rights before choosing tools, because an AI system can only be as safe as the permissions underneath it. They should start with one workflow that has a clear metric, set human approval points for decisions with legal or financial weight, and avoid locking into a single model provider. Our own systems are built to switch models when a better or cheaper one appears.

Leading Practical AI Integration Through Technical Architecture

Bilyal Mestanov is the Co-Founder and CTO of Lexis Solutions (lexis.solutions), the Sofia-based AI integration and custom software development company. He leads the technical direction of Lexis Solutions and designs the architecture of most of its projects, deciding which components an AI-native product actually needs, from data pipelines and retrieval layers to model selection, access permissions and evaluation. He has been the architect behind the company’s most complex and largest-scale work, especially its AI systems, for clients in Germany, the UK, the US and Finland. Bilyal is a Claude Certified Architect, and Lexis Solutions has achieved the Select partner tier in the Claude Partner Network.

Bilyal holds a bachelor’s degree in Computer Science and Technology from the Faculty of Mathematics and Informatics at Sofia University, where he also taught object-oriented programming and data structures and algorithms as a teaching assistant. Before focusing on architecture, he spent years as a lead full-stack engineer, designing and scaling CRM, online reputation, health tech and workplace mental health platforms and leading cross-functional engineering teams. His background in web crawling and natural language processing goes back to his early freelance work. He co-founded Lexis Solutions with three fellow students he met at the HackFMI hackathon in 2016

“The 14-day model at Lexis Solutions delivers one AI-powered workflow in production in 14 days, at a fixed fee between EUR 2,000 and EUR 5,000 agreed before work starts.”