AI Agents•Workflow Automation•Integrations
AI Agents & Workflow Automation
Automate repetitive work with production-ready AI agents.
I design, build and deploy the whole system — from the first workflow map to a service running in production inside your own stack.
- Custom AI agents
- Workflow automation
- CRM & tool integrations
- Production ready
Typical AI projects
Four shapes cover most of the work. Each one is a real system with logging, guardrails and a hand-over — not a prompt in a chat window.
AI Customer Support
An agent that reads incoming questions, answers from your own knowledge base and logs every conversation where your team already works.
AI Document Processing
Invoices, contracts and forms in — structured, validated data out, routed straight into your accounting or back-office system.
AI Research Agent
Gathers information from the web and your internal documents, checks it against sources and delivers a short, structured report.
AI Workflow Automation
End-to-end flows that connect the tools you already pay for, with an LLM only at the steps where it genuinely adds value.
Facturis
AI bookkeeping assistant
Send Facturis an invoice any way it arrives — a phone photo of a receipt, a PDF, or a UBL e-invoice — and it extracts the vendor, amounts and VAT, presents them for a quick review, and books the entry. Every document is routed the smart way: UBL is parsed directly, digital PDFs are read from their text layer, and only scans and photos go through OCR and the language model. It runs hosted in the cloud, or entirely inside your own infrastructure on an open-source model.
- OCR
- AI extraction
- VAT recognition
- Automated booking
- Cloud or fully self-hosted
From workflow to production in four steps
- 1
Discovery
We map the workflow, the tools and what success looks like, then pick one small, high-impact use case to start with.
- 2
Architecture
We decide which parts stay rule-based, where an LLM earns its place, and how data moves between systems.
- 3
Development
I build the agent, backend services and integrations in clean, testable code — usually Python or Node.js/TypeScript.
- 4
Deployment
Validation, guardrails, logging and monitoring, then hand-over with documentation and a roadmap for the next iteration.
AI systems I have shipped
AI Agentic Newsletter
LiveAn autonomous content pipeline: agents scrape eight publications, dedupe stories, curate crypto Twitter and publish a finished newsletter daily — no human in the loop.
HumanityLink
PilotAI-powered onboarding for a humanitarian aid platform, live in pilots with two NGOs in Colombia.
Agent AI Tools
LLM agents for fitness tracking and nutrition guidance, combining a model with data storage and automation scripts.
Built with tools that hold up in production
Model choice is a decision per task, not a religion — and everything can run on your own infrastructure when the data demands it.
Frequently asked questions
Can AI work with our ERP or CRM?
Yes. If your system has an API — HubSpot, Salesforce, Exact, Moneybird, AFAS or an in-house tool — an agent can read from it and write back to it. Where there is no API, we fall back on exports, e-mail or database access.
Can AI run on-premise?
Yes. The full pipeline — OCR, extraction, storage — can run inside your own infrastructure on an open-source model, so no third-party API ever sees your data. That is how the Private Edition of Facturis works.
Do you use OpenAI?
Yes, and Claude, and open-source models such as Llama and Mistral. The model is chosen per task on quality, cost and privacy, and the code is written so it can be swapped without a rewrite.
Can AI work without internet?
Yes. With a local model running on your own hardware, the whole workflow stays inside your network — useful for air-gapped environments and for data that is not allowed to leave the building.
How long does a first project take?
A focused first use case typically runs two to six weeks from discovery to something live. We start small on purpose: one workflow in production beats a roadmap full of ideas.
What if AI is not the right answer?
Then I say so. Plenty of automation problems are better solved with a script, a queue or a fixed rule — cheaper to run and easier to trust. You get an honest read before anything is built.
Let's build your AI workflow.
Book a free consultation. We will map the opportunity and the constraints, and you get a realistic view of what is worth building — no hype.