TLDR
AI that thinks like your best people. Your best people can't be everywhere. Their judgment can. Orbital builds AI systems grounded in how your business actually works: turning the documents a client sends into records a reviewer has checked and signed.
Where AI works in production
Three kinds of system are viable today, in order of increasing autonomy:
- Document intelligence. Reading, classifying and extracting information from unstructured documents at scale: statements, invoices, receipts and agreements.
- Augmented workflow. AI as one step in an existing process, where the AI suggests and a person decides.
- Autonomous agents. A task completed end to end inside set limits, with an escalation path to a person.
All three need evaluation, observability and an escalation path to a person before they run in production. All three use the same models anyone can buy. What makes any of them correct for your firm is expertise: what a right answer looks like, which checks it has to pass, and when to stop and ask a person. The more the system does on its own, the more of that expertise has to be captured before it runs.
A substrate captures your expertise
A firm's captured expertise is its substrate. It has three layers, which are what the firm already knows: quantitative (the metrics and calculations the field uses), subjective (what the metrics miss) and procedural (how an expert responds). Three wrappers are defined from the layers: question archetypes, constraints and safety, and the grounding constraint, which is what every output is tied to. Every answer the AI gives is tied back to both. Worked examples across six professions.
What I provide
Three phases:
- Map. Short sessions with the people who do the work, starting from the artefacts the firm already produces and one real job followed end to end. They write the substrate down: what "complete" and "correct" mean for each job. You receive a practice manual seed, a completeness specification and an architecture note, paid for on its own.
- Architect. The note becomes a design: what goes onto the Document Substrate, which general parts already exist, and what has to be built for you. Your expertise goes into versioned artefacts. One install per customer, in accounts you own.
- Build. The parts that are yours, managed by Orbital, so you pay only for what is yours. Every reviewer correction becomes a permanent test case, and the checks, manual and document types are revised as real jobs show what the first version missed.
The proof
A written substrate needs software to apply it to real work. That software is the Document Substrate: working software, developed on real client work and documented end to end. The inbox is read in any state. Personal details are replaced with placeholders before any model call and restored only for an authorised person, and every restore is logged. Checks run in code: twenty-two document types have field definitions and eighteen have arithmetic checks. The model never does arithmetic, and the workpaper and draft letter are generated from checked records only. Every figure is traced to its page. A person reviews and signs. The Document Substrate in detail.
Workflows fit the Document Substrate to your firm
A workflow defines a job or engagement. It has three parts: slots (the documents the job needs, each named), rules (checks that run in code over the figures read from those documents, with no model taking part) and outputs (a report on screen and a workpaper stamped DRAFT until every slot is filled and every rule passes). The Document Substrate is the same for every firm. The workflows, and the practice manual that explains each rule, are what change from one firm or industry to the next. How Document Substrate works, explained simply.
Principles
Intelligence is bought; expertise is built. Garbage in, garbage out, so validation is the gate. Doer to reviewer, so verification is the accuracy layer. Keep the model out of the parts that have to be defensible. Personal details never leave the building. Ledger, not spreadsheet. One install per customer. Small firms first. Expertise is captured, not generated. Implementation is becoming a commodity; domain knowledge and architectural judgement are what remain scarce. The principles in full.
Who this is for
Small firms and practices whose work depends on documents: accounting and bookkeeping, property management, professional services, councils, co-ops. Also organisations adopting AI that need someone to cover strategy through to delivery on a contract or fractional basis, working alongside business leaders, technical teams and external partners. Not a fit: generic chatbots, problems an off-the-shelf tool already solves.
Who I am
Johan Steenkamp, AI product engineer, over a decade building production systems across fraud detection, industrial monitoring, scientific instruments and geospatial intelligence, as principal engineer, architect and head of software. Available for contract, fractional and advisory work. Christchurch, New Zealand.
Contact: info@orbital.co.nz