Own product
Decree
Legal work is document work — Decree applies AI to it without letting the AI improvise.
Outcome: A working AI product that shows how we keep assistants useful and contained.
Applied AI
AI that does a real job in your business
We build document intelligence, search, assistants and automation for organisations worldwide — on Amazon Bedrock, inside your own AWS account, with the guardrails, evaluation and cost controls that separate a production feature from a demo.
Where AI earns its place
Answers from your own documents
Search and question-answering over contracts, policies, records and manuals — grounded in your documents so the answer can be checked, not invented.
Document processing at volume
OCR and extraction pipelines that turn scanned forms, invoices and applications into structured data your systems can act on.
Assistants that stay in their lane
Customer- or staff-facing assistants scoped to a domain, with rate limits, abuse protection and honest disclaimers — the same way this site's own assistant is built.
Automation inside existing workflows
AI as a step in a process you already run — classifying, summarising, routing — rather than a separate tool nobody opens.
How we keep it trustworthy
- Amazon Bedrock (Claude) in your own AWS account — your data stays yours
- Retrieval-augmented generation so answers cite their sources
- Tightly scoped prompts, rate limiting and abuse protection
- Evaluation before launch and monitoring after it
- Cost controls so a popular feature doesn't become an expensive one
- No medical, legal or financial advice from a model that shouldn't give it
AI we've built
Own product
Moteya Business AI
An AI assistant platform that helps businesses get real work done.
Outcome: An AI platform build for real business use cases.
Reference build
Recipe RAG
A focused build that proves our approach to AI-powered search over documents.
Outcome: A working reference implementation we reuse in client systems.
Questions we're often asked
Will our data be used to train someone else's model?
No. We build on Amazon Bedrock inside your own AWS account; your documents and data stay yours, are processed in-region, and are not used to train foundation models.
How do you stop the AI from making things up?
By grounding it: retrieval-augmented generation so answers come from your documents and can be checked, tightly scoped prompts, evaluation before launch and monitoring after it — plus honest disclaimers where the stakes demand them. This site's own assistant is built the same way.
What does an AI feature cost to run?
It depends on usage, which is exactly why we design cost controls in — rate limits, model choice per task, and dashboards that show spend. The build itself is scoped and quoted per engagement; we don't publish standard prices.
Do we need a data science team to maintain it?
No. We build production AI as software: versioned, monitored, documented and handed over. Your team maintains an application, not a research project.
Is AI compatible with POPIA?
Yes, if it's designed for it: minimise the personal information the model ever sees, process in-region, control access, and log what was asked and answered. We treat POPIA as a design input, not a legal afterthought.
Have a process AI could take off your team's plate?
Tell us about the documents, the workflow and who would use it. We'll tell you honestly whether AI is the right tool.
Talk to an engineer