LMXAI ships AI agents, workflow automation, agentic systems and voice agents — plus private/sovereign stacks when data must stay inside. Traffic-first GEO/AEO pages included — from prototype to production, globally.
# Production agent — tools, evals, optional private serving from lmxai import ServingStack, Agent stack = ServingStack( model = "mdAgent-Hermes-32B", # Qwen3-VL LoRA runtime = "vllm", quant = "awq-int4", gpus = 8, # A100 cluster region = "eu-sovereign", ) agent = Agent(stack, tools=["rag", "mcp", "voice"]) agent.serve(compliance="eu_ai_act") # illustrative API — not a published package
Most AI projects stall between a promising demo and a system that survives production. LMXAI closes that gap with end-to-end ownership.
Software engineering, AI architecture, data science, product strategy and commercialization — delivered by a single accountable partner acting as your fractional AI CTO.
Traffic-facing clusters first: agents, automation, agentic systems, voice, GEO/AEO — with private AI when you need it.
Multi-step agents with tool use, LangGraph orchestration and eval gates — production paths, not chat demos.
Workflow automation that replaces manual ops — document pipelines, RAG jobs, scheduled agents with clear commercial intent.
Voice-capable agents wired to the same tools and policies as your text stack — fast-moving category, same production bar.
Answer-shaped pages and technical surfaces
(llms.txt, sitemap, schema) so agents
and answer engines can retrieve you — not only rank
you.
Niche but high-value: on-prem or private serving when data and weights must stay inside your boundary.
Fine-tunes and compression so tool-use agents fit your hardware budget without losing reliability.
EU AI Act guideConcrete answers on sovereign workspaces, the EU AI Act deliverable, vLLM, production RAG, and the path from prototype to production — in English and Dutch.
Who classifies risk and ships the engineering evidence pack under EU 2024/1689.
Engineering, architecture and product strategy on a single throat to choke.
EU consultancy for LangGraph, tool use and evaluation — not demo chatbots.
When a hosted tenant is enough — and when EU data cannot leave your boundary.
Engineering evidence: inventory, classification, logging, technical file.
GPUs, no-egress networking, and the procurement questions that matter.
Cost, reliability gates, and EU-friendly vendors you can self-host.
Repos, evals and runbooks. Not workshops as a substitute for a system.
The stall points EU teams skip, and the exit criteria LMXAI uses.
A selection of shipped platforms across enterprise, allied health and education.
An enterprise AI workspace positioned against Copilot & ChatGPT Enterprise on EU data sovereignty — keeping company data inside the org's own boundary.
A multimodal tool-use fine-tune of Qwen3-VL-32B, optimized for reliable single-turn function calling and document understanding.
A FastAPI-based LLM gateway with token streaming, full observability and per-user namespace isolation for multi-tenant deployments.
How LMXAI builds reliable production agents — LangGraph orchestration, MCP tool integration, fine-tuned tool reliability (BFCL-v3), observability, and the real workspace tool ecosystem they operate.
A clinical AI platform for dietitians built on LangGraph, automating documentation and clinical workflows for regulated allied-health practice.
A personalized AI study companion for students — a reliable, fast-responding mentor that connects new topics to existing knowledge and supports teachers with content and planning.
Corpus statistics, LDA topic modeling and interactive topic-term visualization — the research layer underpinning Learning Matrix's semantic retrieval.
Where data sovereignty, compliance and reliability are non-negotiable.
Teams shipping agents and workflows — not slide decks — across SMB and larger orgs.
Clinical AI for regulated healthcare and practitioner workflows.
Fintech, insurance, agriculture, customs & logistics.
EU AI Act-aligned, hyperscaler-independent deployments.
I'm an AI engineer focused on agents, automation and private AI based in Leiden, the Netherlands, working at the intersection of LLM research and production engineering. My approach is evidence-driven and skeptical but constructive — I care about systems that hold up under real load, not benchmarks that look good in a slide.
Tell me about your project — fine-tuning, agentic systems, sovereign deployment or AI strategy.
Leiden, Netherlands
info@lmxai.com
