Production AI systems
and custom software,
built in Australia
ComplxAI designs, builds, integrates and operates production AI systems and custom software for Australian businesses. Strategy, engineering and deployment by one team: what we recommend is what we ship.
Six capabilities, one engineering team
Start where the problem is; the same engineers carry it through.
AI consulting
Which problems to solve first, scoped by the engineers who will build them.
- Scored on impact, effort and risk
- Build, buy or defer, in writing
- Board-ready plan
AI development
Working applications where a model does real work on your data.
- LLM applications and copilots
- Document and data processing
- Built on your systems of record
AI engineering
Everything around the model that makes it dependable: retrieval, routing, evals, tracing, hardened AWS deployment.
- RAG pipelines and evaluation harnesses
- Quality, cost and latency control
- Security and data boundaries first
AI agent development
Software that reads, decides and acts across your systems; a person approves what matters.
- Tool use with approval gates
- Multi-step operations, end to end
- Every decision logged
AI product development
Products with AI at the core, from first version to a platform customers pay for.
- A shippable first version
- Multi-tenant SaaS architecture
- Roadmap through to scale
Custom software development
Portals, admin panels, APIs and mobile apps shaped to how you operate, owned by you.
- Web, mobile and admin interfaces
- APIs and integrations
- Handover included
AI strategy
Roadmap and build-or-buy calls.
SaaS development
Multi-tenant products that scale.
AWS & cloud
Serverless, containers, cost control.
AI automation
Agentic workflows for operations.
AI integration
Xero, HubSpot, Salesforce, databases.
AI training
Workshops on your real work.
The full list, with what each one covers: all AI and software engineering services.
What we engineer, layer by layer
Most of an AI system is ordinary software around the model; this is the stack we build against.
- Interfaceswhere work entersCustomer portalsAdmin panelsMobile appsB2B APIsEmail and documents
- Orchestration and agentscontrol flowWorkflow state machinesTool-using agentsApproval gatesQueues and retriesAudit trail
- Retrieval and knowledgecontextDocument ingestionEmbeddingsPostgres with pgvectorManaged vector databasesHybrid searchPermission-aware retrieval
- Models as componentsinferenceModel routerOpen-weight modelsAWS BedrockOpenAI, Anthropic and Google APIsStructured outputs
- AWS platformruntimeLambda and Step FunctionsECS/FargateEventBridge and SQSAPI GatewayS3, DynamoDB, RDS/PostgresCognito and IAMTerraform or CDK
- Evaluation, observability and securitytrustGolden datasetsRegression evals in CITracing per requestCost telemetryCloudWatch alarmsPII handlingTenant isolationPrompt-injection defences
Providers are components, never the product. What matters is retrieval that finds the right passage, routing that controls cost, evals that catch regressions, and a person approving any action with consequences: the substance of our production AI engineering.
Strategy, engineering and production, by the same people
The engineers in the strategy workshop are the engineers on the deploy; the plan only promises what is buildable.
- StrategyThe shortlist, its order and its likely cost, settled by those who will build it.
- ArchitectureWhere data may travel, what runs where, which model does which task, and the contracts between systems.
- BuildThin vertical slices, working software every couple of weeks, evals in CI from day one.
- IntegrateXero, MYOB, HubSpot, Salesforce, email and the databases already in use.
- DeployTerraform or CDK, CI/CD, alarms and a runbook, ready to hand over or take in-house.
- OptimiseQuality, cost and latency tuned on live traces; operated with you or handed over.
Every step is priced in writing first; see how we work for the five stages, or meet the team that does the work.
Three delivered systems,
in numbers
Every card opens its case study.
Modedu, EdTech. Custom SaaS platform, agentic operations and AI-personalised papers; ARR $400k to $1M.
Client-reportedDA Group, NDIS allied health. Onboarding, matching, invoicing and reconciliations on a tenth of the human resource.
Client-reportedLuxpip, FinTech trading. Web and mobile apps, admin panel, B2B API, agentic internal ops; 40% scale-up.
Industries we build for
The system of record first, then the software. Also delivered: EdTech, FinTech, NDIS, media, PropTech and AgTech. All industries. Remote-first across Australia, onsite where it helps: where we work.
Professional services
Client intake, drafting from precedents, billable time recovered.
Retail and e‑commerce
Listings and stock in sync, support that knows the order.
Healthcare and wellness
Referrals, clinical admin and claims, patient data kept inside the boundary.
Real estate and property
Enquiry response, listing copy, tenant and owner admin in the CRM.
Construction and trades
Quotes from plans and price lists, crew scheduling, site diaries.
Hospitality and events
Bookings across channels, guest messaging, rosters that follow demand.
Five stages, priced one at a time
Two free conversations first; then Discovery, Strategy and Implementation, each quoted in writing, and you may stop at any boundary. No mandatory retainers.
See the full engagement modelIntro call
What you run, what is slow or manual, whether we fit.
Pre-discovery
We size the opportunity and price Discovery in writing.
Discovery, Strategy, Implementation
A scored use-case map, a board-ready plan, then software in 12-week blocks.
What buyers ask before the first call
Anything else, bring to the free intro call, or read first: our research on how AI is built and paid for in Australia, insights on agents, automation and the road from demo to production, and the cost and ROI tools.
Ask us directlyDo you build it, or just advise?
Both, by the same people. Strategy ends in a plan we are prepared to build; the engineers who scoped it ship and operate it, or build alongside your team and hand over.
Which businesses do you work with?
Owner-led companies, startups and scale-ups across Australia; the same five stages also run inside an established product team. Common to all: a real problem, usable data, and a decision-maker who wants working software.
Are you going to try to sell me an AI platform?
No. We are independent; nothing we recommend depends on a product we need you to buy. The model behind each task, from OpenAI, Anthropic or Google, on Bedrock or open-weight, is a choice the design keeps open. The code is yours.
Is this only for large companies?
No. Implementation runs in 12-week blocks: a smaller company starts with one system and can stop at any boundary; a larger one runs the same blocks inside an existing product team.
How much does it cost?
Nothing for the first two conversations. After that each stage carries its own written price before it starts, so Discovery, Strategy and each 12-week block is a separate decision. A ballpark comes from the AI project cost calculator before we talk.
What if AI is not the right answer?
We say so in Discovery. Often the answer is plain software with AI in one place, or an integration that removes the manual step. Recommending less is part of the job.
Tell us what should be running
in production by next quarter
A 30-minute intro call, no pitch. We say whether it is worth building and what it would take; if so, a free pre-discovery hour follows and Discovery is priced in writing before anything is paid.