We help mid-sized European companies adopt AI without the consulting bloat — vendor-agnostic, EU-compliance-first, hands-on. Eight years in deep tech, three EU research grants, and two AI products of our own: Makuri (EdTech) and Family Hub (modular AI platform).
Bucharest-based AI consulting practice founded in late 2025. We work with European SMBs (10–200 employees) on AI adoption, AI Visibility, and security audit — vendor-agnostic, EU-compliance-first, and built around real implementation rather than slide decks. Working languages: English, Russian, and Ukrainian. Primary geographic coverage: Romania, Poland, Czech Republic, Germany, and the Netherlands.
We help you go from "we should be doing something with AI" to a working proof of concept in production. Vendor-agnostic — we'll recommend Claude, Gemini, Mistral, or open-source models based on what the problem actually needs. EU compliance built into every step, not bolted on at the end.
When someone asks Claude, ChatGPT, or Le Chat about your company, what do they say? We make sure the answer is accurate, current, and authoritative — through Answer Engine Optimization on your website and a public Model Context Protocol server you own and control.
Threat modeling and security review for AI-powered applications: prompt injection, data leakage, agent misuse, regulatory exposure under the EU AI Act. Available Q3 2026.
AI Visibility is a new distribution layer. As AI assistants become how people search, ask questions, and decide on products and vendors, your company needs to be on that layer — discoverable, accurate, and authoritative.
We didn't just write about it — we built it into our own product, Makuri, before selling the service to anyone. And we built it in both directions: one MCP server that makes the product legible to AI assistants, and a second, authenticated one that lets a user bring their own data into the assistant they already use.
Both are live. Here's each.
A public Model Context Protocol server for Makuri: eleven tools, three reference documents, and two guided prompts. Nine read-only tools cover pricing, languages, subjects, safety, compliance posture, contacts, tech stack, and free resources; two interactive panels — a "How Makuri works" walkthrough and a live Romanian level mini-quiz — run inside the chat. Open source under MIT license. Cold start under two seconds.
An authenticated, per-user MCP endpoint built into Makuri itself. Each student gets a private, read-only, revocable token so their own AI assistant — ChatGPT, Claude, Le Chat — can read their study notebook and revise from it anywhere: "quiz me on my notes", "explain this one more simply", "what should I review today?". Scoped to a single student, with no login, no account access, and no subscription — the notes belong to the student, and so does the right to take them elsewhere. A first, deliberately small step: the architecture is built so external AI agents can talk to the product directly, and that doorway widens over time.
Makuri runs a public Model Context Protocol (MCP) server. Connect it once, then ask your assistant about Makuri — and in ChatGPT it also opens interactive panels right inside the chat, including a Romanian level mini-quiz.
https://mcp.cogniledger.eu/mcpPublic · no login · open source (MIT) · indexed in the Official MCP Registry
The interactive panel is rolling out as Claude finishes its MCP Apps support; the info tools work today.
AI tutoring platform for immigrant children aged 10–16 integrating into EU school systems. Student uploads a PDF textbook or pages photos — gets explanations, translations, and tests in their native language. Two AI modes: fast (Gemini) and deep (Claude Sonnet) for nuanced, context-rich responses. 14 UI languages, parent dashboard, adaptive memory system, config-driven admin panel.
In development: Exam Prep module — preparation for Romanian national exams (Evaluarea Națională + Bacalaureat) with official Ministry of Education scoring barems.
makuri.eu →A configurable AI platform for families. Fifteen live modules — meal planning, language learning, household finances, voice notes, family scheduling, and more — each powered by best-fit AI models routed through a vendor-agnostic gateway (Claude, Gemini, GPT, Whisper, image generation). Built as a private alpha with one Romanian household; one module (SRL Advisor for Romanian small-business owners) is being prepared for a separate commercial release.
A working log of decisions, builds, and tradeoffs from CogniLedger — AI Visibility, public MCP servers, and the day-to-day of running a one-person consulting practice.
Makuri now talks to AI assistants in two directions — outward and inward. That's a deliberate wager on how people will reach products next. Here's what I built, why I started small on purpose, and what it means for any company watching the same shift.
Read the article →What an EU SMB actually gets out of shipping a public MCP server, what to expect over the next quarter, and a three-week playbook to ship your own.
Read the article →Six things bit me when publishing an MCP server to the Official Registry. Each one cost time. Each one is a lesson worth budgeting for next time.
Read the article →AI assistants are becoming a new distribution layer. Most companies aren't on it yet. Part 1 of a three-part series on shipping a public MCP server as an AI Visibility artifact.
Read the article →