INSIGHTS
Implementation notes from real engineering questions.
Read how document pipelines, workflow applications and applied AI can be designed with uncertainty, human review and system boundaries in mind.
Latest writing
Engineering notes, decisions and lessons.
n8n vs LangGraph: When Your Workflow Becomes a Business System
Where n8n works, where LangGraph helps, and why I prefer a dedicated business core for complex rules, recoverable processing and accountable decisions.
Why I want to build Kaliits in Morocco, and share the engineering
My ambition for Kaliits is to reduce paperwork in Moroccan logistics and finance. Sharing OCRAgent and building a commercial offer serve different parts of that goal.
How I want Kaliits to compete in a market full of OCR tools
Human review and document extraction are already available elsewhere. Kaliits has to earn its place through workflow fit, clear scope and evidence from the team's own work.
From an n8n workflow to OCRAgent: why I separated the core
OCRAgent began as an n8n document workflow. Changing requirements pushed me toward hexagonal architecture and a plugin system for different operational needs.
What Kaliits does, explained through one unfinished workflow
A practical explanation of Kaliits: connect the documents, tools and people around one recurring exception, then test whether the change helps.
How Kaliits started after I left Smatch Digital
After resigning from Smatch Digital, I began turning a document workflow into something bigger: OCRAgent, a shared technical core and the commercial idea behind Kaliits.
I went to SIEMA with an OCR demo. I came back with a better question.
The SIEMA conversation that moved my thinking beyond OCR matching: who verifies what arrived, which evidence is trusted, and how a discrepancy gets resolved.
I tried Jev. The interesting part was where I disagreed with it.
A personal note on using Jev to sort public problem signals, overruling a misleading fit score, and leaving the real decisions with people.
OCR workflow plugins: deterministic rules and human review in OCRAgent
How Ironclad separates document-reading providers from business workflows, and why its rule engine and review controls matter more than an impressive extraction demo.
Reliable OCR pipelines: transactional outbox and Redis Streams in OCRAgent
Why reliable document processing depends on committed jobs, recoverable queue delivery and explicit failure states. A source-backed look at the OCRAgent worker architecture.
Inside OCRAgent: evidence-backed OCR architecture for document processing
A code-level walkthrough of Ironclad OCR: how PDF packets become classified documents, traceable facts and reviewable decisions in the engineering behind Kaliits.
How to automate invoice processing without trusting OCR blindly
A practical guide to OCR invoice automation: extraction, validation, exception handling, and reliable handoff to finance workflows.
Modeling receipt and invoice exceptions in the SIEMA prototype
A technical note on the prototype's quantity checks, explicit invoice hold and limits of its scenario data.
Designing document reconciliation around evidence and decisions
An implementation note on separating extraction, field comparison, source ownership and human approval in a controlled document workflow.
Where logistics teams should start with workflow automation
A practical starting point for logistics automation: document intake, incident reporting, warehouse visibility, and controlled operational handoffs.
How to choose an automation pilot that operations teams will actually use
A practical framework for selecting a first AI or workflow automation pilot: one repeated process, clear ownership, measurable outcomes, and safe exceptions.
Different from Kaliits buyer guides
These notes explain architecture, testing and implementation tradeoffs. Kaliits publishes separate buyer-facing guides for import and purchasing workflows.
Explore the project work
See how these ideas appear in selected applications and prototypes.