Client work
What we have built, and for whom
Products our teams work on, named. AI in enterprise software, legal platforms and fiscal automation, plus the frontend and QA work that runs alongside it.
Cloud hosting & infrastructure · Global datacentres An advanced Kubernetes platform that lets customers deploy applications without running DevOps of their own, and AI automation that carries the ticketing and deployment process from the customer's message through to the technical work that resolves it, with a human in the loop.
- Advanced Kubernetes architecture serving customer workloads, with application deployment that needs no DevOps expertise on the customer side
- AI automation across the whole ticketing process, thousands of ticket flows automated: it reads the ticket, works out what is required, and replies to the customer
- The automation does not stop at triage. It executes the technical work needed to resolve the ticket
- A human in the loop on the actions that warrant one, so nothing consequential runs unreviewed
Kubernetes · agentic automation · deployment orchestration · human-in-the-loop
No-code application builder · PerfectApps Inc., United States · A division of Harris Computer Corporation We lead how PerfectApps adopts AI as a Harris business unit: building the AI stack, turning QA into an AI-driven automated pipeline, embedding AI into the product itself, and moving the whole R&D organisation up the AI-DLC maturity curve, with a human always in the loop.
- Help chatbot that answers customer questions about PerfectApps from the product documentation and good-practice rules
- Third-party connection builder: an agent that reads integration documentation and generates the full set of integration parameters, the way Zapier or Make would, but specific to PerfectApps
- Per-client vector databases generated from each customer's PerfectApps data
- An agent that builds working PerfectApps applications from a chat conversation
- Support Co-Pilot: production RAG with nine tools over an MCP-style layer, the serving model chosen against a 600-question evaluation set, human review enforced on every action
- An agent that drives a real browser to reproduce customer-reported bugs and returns a verified trace
- A GPT-3.5 model fine-tuned for the application builder, from dataset curation through evaluation to deployment
- QA turned into an AI pipeline: around 1,200 specs per release, self-healing tests, autonomous triage of failures, and 1,792 Cypress specs migrated during a framework removal
Claude Enterprise · Matcha · Azure Foundry · RAG · MCP · fine-tuning · Cypress, Playwright, Selenium
Insurance agency management · A Harris Computer division, United States A multi-phase, agent-first system, delivered and running in production on their internal systems. Every capability is built as an agent with a registered tool interface, so each phase extends the framework instead of rebuilding it: a RAG assistant over the product manuals, OCR automation for payment processing, live database access through an MCP server, and orchestration on top.
- RAG assistant over the product manuals, with citation tracking so every answer points back to the passage it came from
- Document and chunking agents feeding a pgvector store, with a schema that holds agent memory and state alongside the embeddings
- OCR agent that reads scanned payment requests, extracts and validates the fields with confidence scoring and fallback logic, then creates the request in the existing system
- An MCP server over their MSSQL database, giving agents live query access with schema discovery and schema-aware query validation
- Hybrid search that routes each question to semantic retrieval, the live database, or both, then fuses and ranks results from the two sources
- Natural language to SQL with validation and safety checks, so business users query live data in plain English and get the query explained back
- A master orchestration agent coordinating multi-step workflows, with reporting, visualisation and scheduling agents built on the same framework
- Multi-tenant isolation with row-level security validation and audit logging on every query that reaches the database
FastAPI · PostgreSQL + pgvector · MCP over MSSQL · OCR · text-to-SQL · multi-agent orchestration
Fiscal automation · Romania A fiscal automation platform for the Romanian market, a product of MEDA Business Intelligence, built by our team. Its core is an AI engine for Romanian fiscality that derives the fiscal rules itself rather than having them hand-coded.
- An AI engine that generates the fiscal rules for Romanian tax treatment, built with data science and MLOps techniques rather than a hand-maintained rulebook
- AI OCR that extracts fiscal data from invoices and accounting documents, feeding the rule engine directly
- ANAF integrations, e-Factura and UBL invoicing, and the automated declaration engines (D100, D212, D300, D301, D390, D394)
Data science · MLOps · AI OCR · rule generation · ANAF & e-Factura integration
Legal practice management · GermanyCloud practice-management software that German law firms run their casework on: files, deadlines, beA communication, billing and automation. We contributed to the implementation and brought AI into both the product and the delivery process.
- AI deadline extraction from documents, with the lawyer approving every extracted deadline
- Prompt-driven AI document classification, writing the result back onto the document record
- Agentic workflows connected to case files
- AI-assisted specification and documentation, and automated test generation and maintenance
Document AI · deadline extraction · human-in-the-loop review · workflow automation
Litigation services · United StatesA litigation services platform covering court reporting, remote depositions, record retrieval and complex case management. We implemented the automated extraction and AI interpretation of case material, so attorneys stop reading everything to find the few things that matter.
- Automated data extraction from legal documents, deposition transcripts and medical records
- AI interpretation of what was extracted, rather than capture alone
- Structured summaries generated from the case file
- Case-building support surfaced for the attorney working the matter
Document extraction · transcript and medical record analysis · structured summarisation
Frontend development & QA · Decentralised wireless networkA decentralised wireless network platform. We supplied the frontend application development, with specialist engineers on the build, and owned the QA process running alongside it, so the testing kept pace with what the team was shipping.
- Frontend application development with dedicated specialist engineers
- QA running alongside the build rather than after it
- Release testing before every deployment
- Defects tracked through to closure with the client's team
Frontend engineering · test strategy · release testing · defect management
Quality assurance · SaaSAn SEO tooling platform. Our QA team took ownership of the whole testing process and ran it inside their delivery workflow rather than at the end of it, which is the difference between catching a defect and reporting one.
- Ownership of the testing process end to end
- Test strategy and test case design
- Release testing on every delivery
- Defect management, tracked through to closure
Test strategy · test case design · release testing · defect management
CRM & customer analytics · CRM platformGoPro Fast
Customer analysis over their CRM data.
- Analysis of the customer base held in the CRM
CRM data analysis
DevOps & development · Property management, United StatesA property management company operating apartment communities across several states. We took DevOps ownership alongside development work, cutting infrastructure cost and rebuilding the pipeline the company delivers through.
- Infrastructure cost optimisation
- Build pipeline rebuilt for the way the team ships
- Test pipeline tuned to run inside that delivery workflow
- Application development alongside the platform work
DevOps · CI/CD · cost optimisation · application development