Hours back from document-heavy work.
I build document intelligence and AI knowledge systems for service businesses, reducing the time teams spend finding, reviewing and processing information across complex documents.
- Spending too much time reviewing, searching or processing documents?
- You know AI can help but you aren't sure where to start?
- Already have a prototype that isn't reliable enough for real users?
- Need an engineer who can go from workflow to working software fast?
Turn complex documents into usable knowledge.
Start with one workflow. I handle the design, implementation and integration, and within a week we test a working version with your team.
Document Intelligence
Extract, classify and structure information from PDFs, scans and mixed-format documents. I build OCR and information-extraction pipelines that turn source material into data your team can review and use in existing systems.
ExampleTax statement automation →AI Knowledge Systems
Help your team find and connect information across large document collections. I build RAG assistants, searchable knowledge bases and tool-using agents that retrieve relevant evidence and return answers linked to their sources.
ExampleThe Casemap system →Production AI Engineering
Take a new or existing AI application or automation into day-to-day use. I build the evaluation, integrations and monitoring around it, then deploy it with the checks needed to keep quality, latency and cost visible over time.
ExampleThe CourtSync platform →Hi, I’m Manos.
I’m an Applied AI Engineer with experience delivering AI projects across legal services, publishing, ESG and the public sector. With a background in cognitive science, computational modelling and machine learning, I specialized in intelligent interactive systems during my MSc. That gives me both a technical and a product perspective on building software for daily use.
After more than five years of developing software in Python and building several custom ERP and CRM systems for professionals, I worked as an AI engineer and taught Natural Language Processing at Utrecht University.
Good AI does not replace your team’s judgment. It gives them back the time and headspace to do their best work.
See the case studies →Why work with me?
Building reliable AI products is hard. I set high standards and clear success criteria, aligned with your target KPIs.
Evaluation against real tasks
I use reference datasets, task-specific rubrics, code-based checks and LLM-as-judge scoring to evaluate extraction accuracy, retrieval quality, source-grounded answers and tool use. This gives us a consistent basis for deciding whether a change improves the system.
Simulations and edge cases
I use multi-turn simulations and adversarial tests to check ambiguous requests, missing evidence, conflicting instructions and prompt-injection attempts. Failures become test cases, so the same problems can be checked again as the system evolves.
Guardrails and privacy
I validate inputs and outputs, apply prompt-injection defenses, and detect or redact personally identifiable information where needed. Sensitive information is masked or excluded from application traces and logs, with human review built into decisions that require it.
Tracing and optimization
I trace model calls, retrieval and tool execution to diagnose errors. I compare models and retrieval configurations against the same evaluation set, measuring quality, response time and token cost, and monitor the system after deployment.
What people say about my work
This system helped us tremendously. I cannot imagine the rest of my professional life without this!
I came in skeptical about AI. But Manos built this so every change goes through an editor and the AI doesn't make changes without approval.
What Manos' solution did in a few hours is work that would have taken us months — if we ever did it at all.