Product & Platform Engineering
End-to-end delivery of web and mobile products: architecture, implementation, release, and the operational tooling that keeps them running after launch.
- Next.js & React
- iOS & Android
- Design systems
Crux Coast designs, builds, and operates web, mobile, and data platforms end-to-end. Architecture through to launch, tested, documented, and handed over in a state your team can own.
Our working stack
Capabilities
Six service lines covering the full lifecycle of a production system, from the initial architecture through to the operational tooling that keeps it running.
End-to-end delivery of web and mobile products: architecture, implementation, release, and the operational tooling that keeps them running after launch.
Systems designed for the load they will actually carry. Infrastructure as code, automated deployment pipelines, observability, and predictable running costs.
Pipelines, warehouses, and reporting layers that turn operational data into decisions, with lineage and validation built in rather than bolted on.
Model-backed features scoped to a measurable outcome: retrieval, classification, and workflow automation, with evaluation and human review in the loop.
Connecting the systems a business already runs on. Resilient ERP, identity, and third-party integrations designed to fail safely and reconcile cleanly.
An independent read on an existing codebase: architecture, security posture, delivery risk, and remediation, written for both engineers and decision-makers.
Why teams engage us
Most software problems are not coding problems. They are decisions made early, under uncertainty, that become expensive later.
We run a structured technical review of your existing system covering architecture, delivery risk, security posture, and cost, then give you a written assessment with a prioritised remediation plan. No obligation to engage us to act on it.
Selected work · 2026
Everything below was designed, built, and shipped in 2026, spanning AI products, multi-tenant platforms, business-intelligence and data tooling, native desktop software, and direct-to-consumer commerce we run ourselves.
Platform 2026 Discovery engine · 6,000+ indexed tools
Our flagship platform. A search and comparison engine indexing over 6,000 AI tools, with faceted search across capability and pricing, structured side-by-side comparison, verified reviews, and a moderated submission pipeline for vendor listings.
AI receptionist · Voice & messaging
An AI receptionist that answers inbound calls and messages around the clock, capturing caller intent, qualifying the enquiry, booking appointments straight into a live calendar, and handing off to a human with a full transcript and summary when a call needs one.
Platform 2026 Business intelligence · Xero integration
A no-code BI platform that ingests spreadsheets and live Xero data through an authenticated integration, then models it into executive reporting: revenue, cash runway, and financial-health KPIs.
Quantitative equity screening
A market-research engine that screens an equity universe through three independent scoring models, then surfaces triple-confluence candidates with risk-managed position sizing.
Regulatory compliance · Multi-tenant
A multi-tenant compliance platform for Registered Training Organisations, built around a rules-based compliance engine, an immutable audit trail, role-based access control, and automated notification workflows. Delivered under a private client engagement.
Applied NLP · Communication risk
An applied-NLP application that models how a message is likely to be received before it is sent: tone analysis, persona-based perception simulation, risk classification, and rewrite suggestions.
Native desktop · Windows
A native Windows clipboard manager built on Tauri and Rust: unlimited history, instant full-text search, automatic categorisation, and credential masking. Fully local, with no telemetry or network dependency.
Direct-to-consumer · Retail & wholesale
A direct-to-consumer commerce platform with a full storefront, product catalogue, checkout and payments, plus a separate wholesale ordering channel.
Direct-to-consumer · Devices
An e-commerce platform for clinical-grade therapy devices: catalogue, checkout, and order operations designed, built, and run end-to-end.
Tell us the problem and the constraints. We'll come back with an honest read on scope, approach, and whether we're the right team for it.
Eighteen production modules across finance, workforce, compliance and field operations, replacing a fragmented estate of spreadsheets and email approval chains, while a commercial ERP remains the financial system of record.
Delivered under confidentiality. The client, the product, the sector participants and all personnel are deliberately unidentified, and no screenshots are published, because every production screen carries personal, payroll or client asset data. Specifics can be discussed under NDA.
Showing all 18 modules
Timesheets for a distributed field workforce that survive contact with payroll: line-level approvals, delegated authority, pay-period locking and two-way ERP reconciliation.
Approval is a computed graph rather than a stored list, derived per line from each job's current management chain, so a mid-week reassignment never strands a timesheet.
Leave request, approval and balance management kept consistent with a payroll ERP that is authoritative, opinionated and only partially observable.
Leave entered directly into the ERP was invisible to the platform's matcher. The fix was an explicit reconciliation path built around which system owns which fact, not simply syncing harder.
A visual roster planner for a rotating, site-based workforce: pattern application, bulk editing, camp assignment, fatigue detection and change approval.
A blank cell is a meaningful value: an ordinary working day carries no shift code, so a bulk paint of nulls is an erase. Distinguishing 'no value supplied' from 'value cleared' is essential in any bulk editor.
Access-controlled, auditable employee self-service for payslips and payroll documents, served live from the ERP with no local copy retained.
A permission gate in code is not a permission gate: maintenance scripts were re-granting every role on each data refresh. The durable fix was a never-auto-grant registry enforced by a unit test.
An expense platform that automatically pairs uploaded receipts with corporate-card transactions, codes them, and submits them into an approval chain.
There is no single accuracy number, and inventing one destroys trust. Accuracy is confirmed ÷ (confirmed + rejected), with pending matches excluded from both sides rather than counted as wins.
Client invoice preparation, approval and issue, with rate-card management, contracted-rate leakage analysis and document-management filing.
Most 'unreliable integration' complaints are really error-message problems. Translating an opaque filing failure into an actionable file-locked conflict removed a recurring support burden outright.
A reporting layer over an ERP general ledger making job-level cost, revenue and margin visible without waiting for month-end.
The nightly sync re-reads a trailing window because postings arrive late, and is read-only by construction: it issues GET requests exclusively, so a reporting bug can never corrupt the ledger.
Purchase requisition through to ERP-posted order, with delegated-authority routing, goods receipting and per-file attachment chaining.
Retry has to know what already succeeded. A failed post and a successful post with failed attachments are different states. Blindly repeating the operation would duplicate the order in the ERP.
Supplier onboarding and approved-vendor management, with compliance verification, insurance-currency tracking and international banking validation.
Validation rules must match reality, not intuition. A 9-digit account-number cap assumed a domestic format and silently blocked every legitimate overseas supplier sharing that field.
A human-resources data layer over a payroll ERP, adding the directory, org charting, onboarding and reporting the ERP does not provide.
A termination date is not proof of termination. Records where it equalled the start date were data-entry artefacts. The reliable signal was recent timesheet activity, not a single field.
Safety-critical training compliance: mapping every role to its required competencies, tracking each employee against them, and surfacing the gaps geographically before deployment.
The competency matrix defines what 'qualified' means, so changing it is a controlled act with its own approval workflow and segregation of duties, not an ordinary edit.
Automated verification of training certificates against the issuing authority's own records, confirming the qualification exists and belongs to the person named.
Anything unrecognised is reported as unreadable, never as not-found. Both outcomes are assertions about a real person's qualifications, and a false accusation is far worse than a missed check.
An in-house LMS replacing a commercial SaaS platform: authoring, publishing, delivery, assessment and completion tracking, wired into the competency framework.
An enrolment ID is not an authorisation. Resolving the learner from the enrolment alone lets a caller file a pass against somebody else, so identity is checked against the enrolment's owner.
An in-application assistant answering user questions from the system's own source code and documentation, with no curated knowledge base to maintain.
Lexical retrieval was chosen over vector embeddings on measured grounds: for a corpus with a consistent internal vocabulary, exact identifier matching beats semantic similarity, and removes an entire pipeline.
A field inspection platform implementing a formal international standard: equipment registers, statistical sampling, graded inspections, defect management and offline-first mobile capture.
Offline is a data-model problem, not a caching problem. Conflict is a first-class, auditable state with server version, local version, conflicting fields and an attributed resolution.
End-to-end test-and-certify management for serialised safety-critical equipment, plus a fleet-level reliability analytics layer over the accumulated records.
The headline pass/fail is derived from the test data at read time rather than entered separately, eliminating the defect class where the summary and the detail disagree, with no risky migration.
Assembly and verification management for pressure-system joints: registers, torque calculation, hold points, work packs and field verification.
A hold point is modelled as an explicit workflow primitive rather than a status field, because the audit question is who released the gate and when, not merely what state the record reached.
Chain-of-custody tracking for equipment moving between site, transport and workshop, with QR identification, signed dockets and throughput analytics.
Movement events are high-frequency, and per-event email trains users to ignore them. Configurable rules feed a digest instead, so the notification stays worth reading.
Cross-cutting foundations
Shared platform capabilities rather than per-module implementations, the layer that determines whether an integrated system is trusted in production.
Engineering standards
Good engineering is verifiable. These are the standards every engagement is held to, and the evidence you receive that they were met.
Automated tests, type safety, and continuous integration gate every change. Nothing reaches production on the strength of a manual check alone.
Architecture decisions, runbooks, and handover documentation ship with the code. Every engagement is designed so another team could pick it up.
Logging, error tracking, and performance monitoring are in place before launch, so regressions surface as data instead of support tickets.
How we work
A deliberately unremarkable process. The value is that each stage produces something you can review, question, and sign off before the next one starts.
We define the problem, the constraints, and what success has to look like in measurable terms before any code is written.
A documented technical design: data model, system boundaries, integrations, and the trade-offs behind each decision.
Short iterations against a working build. You see functioning software early and continuously, not a demo at the end.
Launch, monitoring, and iteration. Support arrangements are agreed up front so ownership is never ambiguous.
About
We build our own products and we build for clients, and we hold both to the same standard, because we are the ones who have to operate them.
Crux Coast covers the full stack of a production system: interface and interaction design, application and API engineering, data modelling, cloud infrastructure, and the release and monitoring tooling that surrounds it. Work is delivered in-house rather than subcontracted, so the people who designed the architecture are the people who build and support it.
Our portfolio spans a published iOS title, multi-tenant compliance and business-intelligence platforms, applied-AI applications, native desktop software, and commercial storefronts we run ourselves. That range is deliberate: it means our recommendations come from having operated these systems, not just specified them.
Common questions
Greenfield product builds, platform rebuilds, enterprise integration work, and technical due diligence. Engagements typically run from a scoped four-week build through to ongoing platform ownership.
Clients own the code, the infrastructure, and the documentation outright on delivery. Repositories, cloud accounts, and pipelines are handed over in a state another team can operate.
Fixed-scope pricing for defined deliverables, and monthly retainers for continuous delivery or platform ownership. Both are quoted after discovery so the number reflects the real problem.
Regularly. We integrate with in-house teams as an additional delivery stream, take ownership of a specific subsystem, or provide architecture and review support alongside your own engineers.
Enterprise engagements are usually covered by confidentiality terms. Where that applies we describe the engineering: the architecture, the constraints, and the outcomes, without identifying the client, the sector participants, or any individual. We can discuss specifics under NDA.
We are based in Australia and work remotely with clients across Australian and international time zones, using written specifications, tracked issues, and scheduled review calls.
Contact
Send through the problem, the constraints, and where it currently stands. You'll get a considered reply from an engineer within two business days, not a sales sequence.