Ship Keys
Launch-readiness checks for apps built with AI coding tools, delivered as a Claude connector and a Claude Code plugin, with autonomous agents running in customers’ CI pipelines. Private beta at shipkeys.ai.
shipkeys.aiMost of my client work is private. These are the systems I can talk about: a live UK structured-products investment platform, seven years of technology leadership at a £1bn+ advice firm, and the products where the agentic AI patterns were worked out.
Designed, built and now operates a UK structured-products investment platform for retail investors and their advisers. Delivered in under a year and live on the web with its first investors.
The estate. When I took it on as Head of Software Development, the firm ran 44 applications and 45 databases, built over years in whatever was current at the time: Visual Basic, VBA, C#, Angular, and later TypeScript and React, on SQL Server and MongoDB, across multiple virtual machines, alongside third-party systems. I took charge of all of it, from the servers and databases to the code, and brought it together in stages: first into two core systems, then into one unified platform. Staff worked in one place instead of many.
Leadership and adoption. I led a technical team of five, reporting to the managing director, and owned software, infrastructure, security, backup and disaster recovery, suppliers and data. I supported more than 80 staff through the adoption of new platforms. I also served as Deputy Data Protection Officer, with GDPR Foundation / Practitioner training from IT Governance (2022).
What I built. A React investment onboarding platform rolled out to more than 80 staff including 18 advisers, which cut application processing time by over 90% and removed paper applications entirely. A direct-to-consumer digital platform. AI-driven portfolio analysis tools. Migrations to modern virtualised infrastructure.
Client reporting. Accountable for six-monthly client valuation reports and annual costs and charges reporting for over 4,000 clients, ensuring accuracy before issue.
The exit. After the firm's multi-million-pound sale, I led the Lowes post-sale integration into the acquiring group: data migrated, overlapping platforms merged, legacy systems decommissioned. I was retained afterwards as a consultant by the managing director who sold the firm.
I build my own products with the same tools I recommend. They are where the approval flows, MCP servers, voice agents and guardrails were first proven.
Launch-readiness checks for apps built with AI coding tools, delivered as a Claude connector and a Claude Code plugin, with autonomous agents running in customers’ CI pipelines. Private beta at shipkeys.ai.
shipkeys.aiA phone and web voice agent with multi-agent hand-off and secured actions. Built on Pipecat and Google Cloud Run, with MCP clients for tools.
Persistent goals and context across conversations, with owner-approved actions, per-call metering and spending limits. End-to-end voice interaction through the MCP connector. Private beta.
Daily records for families of children with additional needs, summarised by AI, with strong privacy controls and model choice by query complexity. In development.
Weekly AI competitor reports for small businesses, with owner approval before anything is published or sent. Prototype.
Review-pack drafting with RAG and LangGraph pipelines, human approval checkpoints and evals, built on synthetic data. Prototype.
The habits came from places where software failure had physical consequences.
Ask on the first call. I am happy to walk through the architecture, the controls and what I would do differently.
steve@minibit.co.uk · Newcastle upon Tyne, working across the UK