Technology
Technology, security and privacy.
The technology is not the product. A workflow that runs is. Every system we build uses the same proven layers, controls and tools. That is what makes it dependable, transferable to your team, and auditable when it matters.
Principles
Six rules we build by.
Model-agnostic
We use the model that fits the job and your constraints: Anthropic Claude, OpenAI or Google Gemini, or open models when data must stay on your own infrastructure. Workflows can switch models without a rebuild.
Yours by default
Accounts, data, credentials and cloud subscriptions sit in your company’s name wherever the vendor allows it. We work in them under a written access policy.
Integrate before replacing
We connect to the ERP, POS, accounting, CRM and messaging tools you already pay for, and replace a system only when the business case says so.
A person where it matters
Anything irreversible, sensitive or regulated has an approval step and an audit trail.
Privacy and security by design
The Data Privacy Act, the National Privacy Commission’s guidance on AI and your own policies shape the design from the first workshop.
Measured
Every workflow reports its own volume, time saved, error rate and running cost.
Architecture
One system, six layers.
You see one system. Underneath, every solution is assembled from the same six layers, with security, identity and monitoring running across all of them.
- Layer 1
Channels and sources
Where work arrives and where data lives
Email, web forms, Messenger, Viber, SMS, phone, POS, ERP, accounting systems, spreadsheets, scanned documents
- Layer 2
Integration
Connects systems and cleans up the data between them
APIs, webhooks, secure file transfer, database connectors, the BIR electronic invoicing interface where required
- Layer 3
Orchestration
Runs the workflow: routing, rules, retries, approvals and schedules
Workflow platforms such as n8n and Microsoft Power Automate, queues, scheduled jobs
- Layer 4
AI services
Reads, classifies, extracts, drafts, summarises and answers from approved knowledge
Large language models by API, document AI, speech to text, search over your own documents
- Layer 5
Data and records
Keeps a durable record of every input, decision, output and human override
Managed PostgreSQL, a data warehouse such as BigQuery, object storage, audit logs
- Layer 6
Experience and insight
What your people use and see
Web and mobile applications, portals, dashboards, alerts, chat interfaces, management reports
Across every layer: single sign-on and role-based access, encryption in transit and at rest, secrets management, logging, and uptime and cost monitoring.
Stack
The tools we standardise on.
Standard tools keep delivery fast and support predictable. Where you already run a platform, we build on it.
| Area | What we use |
|---|---|
| Cloud | AWS and Microsoft Azure, or Google Cloud where you already are. Singapore regions for low latency, your own tenancy when you have one. |
| AI models | Anthropic Claude, OpenAI and Google Gemini through their business APIs; open models for on-premise cases. |
| Automation | n8n, hosted in your environment, and Microsoft Power Automate for Microsoft 365 companies. |
| Applications | TypeScript, React and Next.js for web; React Native for mobile; Python for data and AI services. |
| Data and dashboards | PostgreSQL, BigQuery or a similar warehouse, and Metabase or Power BI dashboards your team can extend. |
| Messaging | Direct integrations with Messenger and Instagram, Viber Business, SMS gateways and email. |
| Payments and invoicing | Local payment rails such as PayMongo and Xendit, and integration with the BIR Electronic Invoicing System. |
Governing AI
AI that is useful, and kept in check.
AI components are capable and fallible. The National Privacy Commission expects privacy impact assessments, privacy by design, monitoring and human intervention for AI that handles personal data, and it has said that using a third-party AI system does not transfer your liability. These controls are standard in every workflow we build.
| Control | How it works |
|---|---|
| Grounding | Answers about your business come from an approved, versioned knowledge base, not open-ended generation. |
| Boundaries | A written list of topics the system always hands to a person, such as clinical, legal or financial advice, complaints and exceptions to policy. |
| Human approval | Required before anything irreversible: payments, refunds, cancellations and commitments to customers. |
| Data minimisation | Personal identifiers removed before a model call whenever the task does not need them. |
| Audit trail | Every request, response, action and human override recorded with the time and the person or system responsible. |
| Testing | A set of real, anonymised test cases for each workflow, run again before any change to prompts or models. |
| Manual switch | One control that returns the workflow to manual operation, tested at handover. |
Sources: National Privacy Commission, Advisory Guidelines on AI (December 2024); NPC statement on third-party AI, July 2026 (MLex).
Security and privacy
Your data stays yours.
What we do on every engagement. We do not claim certifications we do not hold; when a formal certification is in place, it will be listed here.
- Individual accounts, least-privilege access and multi-factor authentication on every system we operate.
- Encryption in transit and at rest, with client data kept in the client’s own tenancy wherever possible.
- Confidentiality agreements with everyone who works on client systems.
- A data processing agreement with every client whose personal data we handle.
- A breach response plan that includes the Data Privacy Act’s notification duties.
- Retention and deletion schedules agreed with each client, and access removed at the end of an engagement.
Security questionnaires
If your procurement or IT team has a vendor security questionnaire, send it with your consultation request. We answer it in writing, with architecture diagrams for the proposed system.
Read our privacy notice
Let’s make work work better.
Your people’s time, back.
It starts with a conversation.
Tell us about the work that slows your business down. Leave your free consultation with a written summary and a practical next step.
Book a free consultationA conversation. A clear next step. No obligation.