What we build

One way of working, whatever the work is. Every engagement is scoped and priced in writing before it starts, built where you can watch it, and handed over with the keys: code, prompts, evaluation sets and all.

Web and commerce platforms

Marketing sites, customer portals and headless commerce that stay fast on a phone connection.

At handover: a deployed site or storefront, its content model, and the pipeline that ships changes to it

Try a working demo

Web development

Web platforms and marketing sites that stay fast on a phone connection and readable by a screen reader.

Current frameworks, a content model your team can operate without calling us, and a performance budget the build fails when it breaks. Accessibility to WCAG 2.2 AA is part of the build, because retrofitting it afterwards costs more than doing it once. This site is a fair sample of the standard.

What is included

Web platforms, marketing sites and content models
Server rendering, caching and Core Web Vitals budgets
Accessibility to WCAG 2.2 AA, verified with axe-core
Structured data, sitemaps and technical SEO
Analytics, forms and third-party integrations

Search and AI visibility

Pages a search engine can index, an assistant can quote, and a buyer can act on.

Two audiences read a website now. Search engines still send most of the visitors, and assistants increasingly read the page and answer on the reader’s behalf. Both reward the same work: pages that load fast, structure that says what a thing is, facts marked up so a machine can read them, and writing that answers the question somebody actually typed. We build that into the site, plan the pages with you, and run the campaigns that point at them. What we will not do is promise a position in a result page or an AI answer, because nobody can.

What is included

Technical SEO in the build: speed budgets, structured data, sitemaps, crawlable markup
Page structure and content planned around the questions buyers ask
Facts written to be quotable: plain answers, clear headings, machine-readable detail
Reporting that names what changed and what it did
Paid, email and social campaigns pointed at pages built to receive them

Mobile and connected products

iOS and Android from one codebase, and the hardware that reports back to them.

At handover: signed iOS and Android builds, both store listings, and an automated release pipeline

Try a working demo

Mobile app development

React Native apps for iOS and Android from one codebase, with the release pipeline set up before the first build.

Offered as capability rather than portfolio. We have the engineering for it and would be glad to be asked. What we will not do is point at somebody else's app and imply it was ours. One React Native codebase serves both stores. Native modules go in where the platform genuinely differs. Store submission is automated, so a release is not one person following a checklist at midnight.

What is included

React Native for iOS and Android from one codebase
Offline handling and background sync
Push notifications and deep linking
Automated builds and store submission
Crash reporting and release monitoring

IoT development

Devices reporting to somewhere useful: telemetry, provisioning, and the dashboard the readings actually land on.

Also a capability offering, on the same terms. The interesting part of IoT is rarely the device. It is what happens to the readings: how a fleet is provisioned and updated, how telemetry survives a lost connection, and what the alert threshold is before somebody drives out to a site. That is a data and delivery problem, which is the work above pointed at hardware.

What is included

Device telemetry ingestion and time-series storage
Fleet provisioning and over-the-air updates
Edge buffering for intermittent connections
Alerting thresholds and escalation rules
Operator dashboards and historical reporting

Enterprise operations

The systems operations actually run on: stock, orders, logistics, and the internal tools around them.

At handover: connected workflows, role-based access, audit trails, and the dashboards operations run on

Try a working demo

Workflow automation

Closing the manual gap between two systems that were never designed to talk to each other.

Approvals, enquiry routing, document handling, reconciliation, reporting. We map the process as your team actually runs it, which is rarely how the org chart describes it, and then automate it. AI goes in only where the step needs judgement. Ordinary code does the rest, faster and for less.

What is included

Process mapping and automation candidate analysis
Integration across the systems you already pay for
Human-in-the-loop review at the consequential steps
Deterministic logic wherever AI is not warranted
Monitoring, alerting and failure handling

Custom software and internal tools

The line-of-business systems the AI has to live in, and the integrations between them.

This is still a good part of what we do, and it is often the right first engagement. An AI feature bolted onto a business that cannot deploy reliably is not an AI problem. We build the internal tools the work actually runs on: the approval screen, the review queue, the admin panel nobody demos but everybody uses.

What is included

Internal line-of-business tools and admin panels
API and third-party integrations
Migration off spreadsheets and shared drives
Role-based access and audit trails
Documentation and team handover

Supply chain and operations systems

Operational software connecting inventory, purchasing, orders and shipments across the systems already in use.

Operational software connecting inventory, purchasing, orders and shipments across the ERP, accounting and carrier systems already in use. We begin with the highest-friction workflows, adding approvals, traceability, dashboards and exception alerts without forcing a risky full-platform replacement.

What is included

Inventory, order and shipment tracking
Integrations with ERP, accounting and carrier systems
Approvals, audit trails and role-based access
Operations dashboards and exception alerts
Migration off spreadsheets and legacy tools

Security engineering

Security built into what we ship, a review of what you already run, and the evidence your customers ask for.

We build with least privilege by default, secrets kept out of the code, dependencies watched and an audit trail that says who did what. On a system you already run, we assess what is exposed and fix it: permissions that grew over time, data reachable by more people than intended, infrastructure left open because it was quicker. Where a customer or a regulator is asking, we gather the evidence and write the answers, which is usually the part nobody has time for.

What is included

Access control, secrets handling and dependency hygiene in every build
Review and hardening of systems already in production
Permission boundaries for agents, and a record of what each one touched
Readiness for security questionnaires, ISO 27001 and SOC 2 evidence
Data protection mapping: what is collected, where it goes, who can see it

Data and cloud delivery

Pipelines, warehousing, access control, and the release machinery that puts a change live safely.

At handover: data pipelines, a warehouse, access controls, and deployments with monitoring attached

Try a working demo

Data foundations

Pipelines, warehousing and access control, which is usually what has to happen before any of the rest works.

Roughly half the AI projects we are asked about turn out to be data projects first. Records live in four systems that disagree. Nobody owns the definition of a customer, and the reporting is a monthly export into a spreadsheet. We fix that layer: consolidated sources, modelled tables, quality checks that fail loudly, and access rules that let you say yes to a request without a meeting.

What is included

Source inventory and ingestion pipelines
Warehouse or lakehouse setup
Transformation and data modelling
Quality checks and freshness monitoring
Access control, lineage and governance

Delivery platforms and DevOps

Harness.io pipelines, feature flags and deployment automation, so what we build reaches production repeatedly.

CI/CD on Harness.io, environment and secrets management, and progressive delivery behind feature flags. We rehearse the rollback before you need it. AI systems change faster than ordinary software: prompts move, models move, retrieval configs move. The path to production ends up being the bottleneck far more often than the code.

What is included

Harness.io pipeline design and setup
Feature flags and progressive rollout
Environment, secrets and access management
Automated test and evaluation gates before release
Rehearsed rollback and release runbooks

Running it after launch

The arrangement for after handover, for teams who would rather not carry it themselves.

Ownership does not change: the code, the prompts, the evaluation sets and the accounts are yours at final payment either way. This is about who operates it. We keep the evaluation runs on a schedule, move prompts and configs as models move, patch what needs patching, and answer within an agreed time when something is wrong. Hosting can sit in our accounts or in yours with us on call, and feature work continues against a monthly allocation you approve.

What is included

Retained engineering: scheduled evaluation runs, model and prompt upkeep
A named response time during agreed hours when something breaks
Hosting and infrastructure in our accounts, or in yours with us on call
Ongoing feature work against a monthly allocation you approve
Monitoring, alerting and a rehearsed rollback

AI and automation

Assistants that cite their source, agents with a job description, document intelligence.

At handover: an evaluation set, the accuracy scores it produces, and the service deployed behind an interface

Try a working demo

AI assistants and RAG

Assistants that answer from your own documents and show which document the answer came from.

Ingestion and chunking, embeddings, hybrid keyword and vector search, reranking, and a citation back to the source. We then measure it against real questions from your staff. A retrieval system that scores well on a public benchmark and badly on your invoices has told you nothing worth knowing. Often the first useful finding is that the documents need work before the model does.

What is included

Source audit, ingestion and chunking strategy
Vector store selection: pgvector, Qdrant or managed
Hybrid search with a reranking pass
Citations and source attribution in every answer
Evaluation sets built from your staff's real queries

Agent automation

Agents that carry out multi-step work inside permissions you set, and leave a record of what they did.

Coding and operational harnesses connected to your systems through MCP servers: Claude Code, Codex, or a custom runtime. On top of that, routing, shared state and handoffs between specialised agents. We define what each agent may touch, and which steps stop for a human. Tracing goes in as well, so you can watch a long run while it happens rather than piecing it together from logs the next morning.

What is included

Harness selection, configuration and prompt architecture
MCP servers and tool integration against your systems
Routing, shared state and handoffs between agents
Permission boundaries and human approval gates
Tracing and run observability

Document intelligence

Turning invoices, purchase orders, contracts and forms into structured data your systems can act on.

Most businesses have someone retyping numbers from a PDF into an ERP. We replace that with extraction that reports its own confidence. High-confidence records post automatically. The rest go to a review queue, where a person confirms them in seconds instead of keying them from scratch. Accuracy is measured per field, so you can see exactly where it is safe to stop checking.

What is included

OCR and layout parsing for scans and photographs
Extraction schema design per document type
Validation rules and per-field confidence thresholds
Human review queue for low-confidence records
Posting into your ERP, accounting or line-of-business system

Customer support automation

Answering the repeat questions, and routing the rest to the right person with the context already attached.

A support inbox is mostly the same forty questions. We ground answers in your help material so replies cite a source, set the confidence bar for what goes out automatically, and escalate everything else with a summary and the customer's history attached. Deflection is reported honestly, against tickets actually resolved rather than tickets merely answered.

What is included

Intent classification and routing
Grounded reply drafting with citations
Escalation rules and clean handoff to a person
Integration with your helpdesk or shared inbox
Deflection and satisfaction measurement

The same four phases, every time

ClarifyWeek 1
PlanWeeks 2-3
Build and verifyWeeks 3-7
Learn and adjustWeek 8 onward

Payments

40%to commence
30%on design approval
30%on completion

A typical six to eight week build. The dates and the price for yours are in the proposal before anything starts.

Discuss a problem How the work runs

Questions we get asked first

Read the questions