Diagnostic
Two weeks. Workshops, a data and systems audit, and a written recommendation with costed options. Fixed fee, no obligation to continue.
Services
Four practices,
one accountable team.
Each practice is staffed by specialists and run on the same delivery process. Below is what each one actually produces: the artefacts you keep when the engagement closes.
Training & mentorship lives here01 · Technology Solutions
Software that
holds up in production.
Product builds, internal tools, integrations and the infrastructure underneath them. Built the durable way: version control, automated tests, containerised deployment and separate environments.
Typical engagement: 6–20 weeks, or a retained team.
- /Web & mobile applicationsCustomer-facing products and internal platforms, built responsive and accessible from the first commit.
- /Operations management platformsEnd-to-end systems that run business operations: records, workflow and approvals, billing, staff, scheduling and reporting. Delivered across sectors including education, where that means enrolment, results and fees.
- /Systems integration & APIsMaking software that was never meant to talk to itself exchange data reliably: payments, ERPs, CRMs, government and third-party services.
- /Cloud architecture & DevOpsProvisioning, containerisation, CI/CD, environment parity, and cost that scales with usage rather than surprise.
- /Process automationReplacing the spreadsheet-and-email workflow with something auditable, including document handling and approval chains.
- /Legacy modernisationIncremental replacement of systems that still work but can no longer change. Strangler-pattern migrations, not big-bang rewrites.
- /Technology consultingAI-readiness assessments, digital transformation roadmaps, automation strategy, stack recommendations and change management, so the tooling decision is made before the money is spent.
- /Support & maintenanceDefined response times, patching, dependency hygiene, and a documented escalation path.
Usually built with
TypeScript · React & Next.js · Node.js · NestJS · Python · PostgreSQL · Prisma · Docker · Vercel · AWS · React Native
02 · Artificial Intelligence
Applied AI with
a job to do.
We start from a task someone currently does by hand, then decide whether a language model belongs anywhere near it. When it does, we build it with evaluations, guardrails and a fallback for the day the model is wrong.
Typical engagement: 4-week pilot, then production build.
- /AI opportunity assessmentA short, unsentimental review of where AI would pay for itself in your operation, and where it would not.
- /Assistants grounded in your knowledgeRetrieval-augmented systems that answer from your documents, policies and records, with citations back to the source.
- /Document & form intelligenceExtraction, classification and validation for invoices, statements, applications, contracts and scanned records.
- /Natural language processingChatbots and voice assistants for customer interaction, sentiment and intent classification, summarisation of long-form records.
- /Recommendation enginesTailoring products, content and next-best-action to observed behaviour, to lift engagement and retention.
- /Agentic workflowsMulti-step automations that call your own tools and APIs, with human approval at the points that carry risk.
- /Evaluation & monitoringTest sets, scoring rubrics, regression checks and cost tracking, so quality is a number, not a feeling.
- /Responsible-use guardrailsData-handling boundaries, prompt-injection hardening, redaction, audit logs and a documented escalation path to a human.
Usually built with
Anthropic Claude · LangChain & LangGraph · FastAPI · pgvector & ChromaDB · sentence-transformers · pypdf & python-docx · Python
03 · Machine Learning & Data Science
Models that
survive contact
with reality.
A model is only useful if it holds on data it has never seen, in a system someone maintains. We validate honestly, deploy properly, and monitor for drift from the day it goes live.
Typical engagement: 8–16 weeks including deployment.
- /Predictive modellingChurn, demand, credit and collections scoring, propensity, lifetime value, framed against the decision it will inform.
- /Time-series forecastingSales, inventory, energy and capacity planning, with prediction intervals rather than a single misleading number.
- /Anomaly & fraud detectionBehavioural and transactional outlier detection tuned to the false-positive rate your operations team can actually absorb.
- /Segmentation & clusteringCustomer and portfolio segmentation that produces groups your commercial team recognises and can act on.
- /Feature engineering & experiment designReproducible pipelines, leakage checks, holdout discipline, and A/B design for measuring lift in the real world.
- /Deployment & monitoringModels served behind an API, with experiment tracking, versioned runs, and performance watched after release so drift is caught before it costs anything.
Usually built with
Python · pandas & NumPy · scikit-learn · XGBoost & LightGBM · PyTorch · statsmodels & Prophet · MLflow · Jupyter · FastAPI · PostgreSQL · Docker
04 · Data & Analytics
One number,
one definition.
Most reporting problems are not chart problems. They are definition problems. We build the pipeline and the shared vocabulary first, then the dashboards on top become genuinely trustworthy.
Typical engagement: 4–12 weeks, often the first project we do together.
- /Data engineering & pipelinesIngestion from operational systems, files, APIs and third parties, orchestrated on a schedule, monitored, and re-runnable without fear.
- /Warehouse & modellingA dimensional model with tested transformations and a metrics layer, so "active customer" means the same thing in every report.
- /Dashboards & reportingExecutive and operational views designed for a decision, not for decoration. Accessible colour, honest axes, no vanity metrics.
- /Data strategyDeciding what is worth tracking, how it will be measured, and what action each number is supposed to trigger, before anyone builds a chart.
- /Data cleaning & preparationStructuring, deduplicating and validating messy source data so the numbers downstream survive scrutiny.
- /Reporting automationRepeatable, scheduled reporting that removes the monthly scramble and the copy-paste errors that come with it.
- /Data quality & governanceValidation rules, freshness checks, lineage, access control and a documented owner for every dataset.
- /Analysis & diagnosticsCohort, funnel, retention and unit-economics work, answering a specific question and writing down the method.
- /Migration off spreadsheetsLifting the reports the business already depends on into something versioned, permissioned and automated.
Usually built with
PostgreSQL · SQL · Python · dbt · Airflow · BigQuery & Snowflake · Power BI · Tableau · Looker Studio · Excel
05 · Mentorship & Training
Capability,
transferred.
Our fifth practice runs as Loreon Academy: cohort tracks, one-to-one mentorship, and private training for teams. It is also how we close every build engagement.
- /Cohort tracksData science & machine learning, data analytics & BI, and software engineering, taught by engineers who ship for clients.
- /One-to-one mentorshipFortnightly sessions, code and portfolio review, and a plan aimed at a specific role rather than a certificate.
- /Corporate & team trainingPrivate workshops built on your own stack and data: SQL, Python, analytics, applied AI, engineering practice.
- /Handover trainingIncluded with every build: runbooks, architecture walkthroughs and working sessions until your team is comfortable operating the system.
Engagement shapes
Three ways
to start.
Scope and commitment scale with confidence. Nobody should sign a twelve-month contract with a team they have not worked with yet.
Project build
A defined outcome with a defined budget and date. Fortnightly demos, a named lead, and documentation shipped alongside the code.
Retained team
An embedded squad for a quarter or more: engineering, data and analytics capacity you direct, with our review practices attached.
How we build
Practices you can
see in the code.
Four habits that show up in every repository we own. They are described here because they are already true, rather than as a standard to be held to later.
Automated tests
Test suites live alongside the application code and cover the parts where a silent failure would cost something: scoring, scheduling, rate limiting, authentication.
Separate environments
Development and production are defined as distinct, containerised configurations, so what runs in front of your users is not the machine somebody happened to be working on.
Secrets stay out of the repository
Every project ships an environment template rather than real credentials. Keys live in the deployment platform, and rotating one does not mean editing source code.
Boundaries enforced by the database
Where one system serves several organisations, isolation is enforced in the database rather than in application code, so a query that forgets to filter returns nothing at all.
Next step
Bring us the
awkward problem.
The ones where nobody is quite sure what the data says. Those are the engagements we do our best work on.