Kyle Doidge

• Built the retrieval-augmented generation (RAG) layer the product runs on, covering chunking, embedding and semantic search over a user's notes, so the assistant answers from the user's own corpus rather than model priors and writes back into it.
• Designed an LLM agent pipeline on Anthropic's API with tool use, multi-step orchestration and structured output, served from a Go backend to a React Native client.
• Fine-tuning a Qwen-family open-weight model to run complex note-processing tasks at a fraction of frontier-model cost and latency, evaluated against a task set built from real usage traces.
• Made a non-deterministic agent pipeline reliably regression-testable in CI, using a deterministic fake provider alongside testcontainers-backed integration tests with per-test user isolation.
• Full Detox end-to-end coverage on React Native, chosen over Maestro after evaluating both on debugging ergonomics and CI stability.

02.1Royal Navy / MoD — AI Ethical Assurance Platform

• Joined a stalled project with no agreed product direction and a fixed delivery date; produced the Figma design system and end-to-end flows that became the team's single reference point, unblocking parallel work across four workstreams.
• Led frontend delivery as the only dedicated frontend engineer, designing and building a bespoke SvelteKit design system of 40+ accessible components and then shipping the full application on top of it.
• Built the platform's core capability, including a user-configurable assurance workflow and dynamic form engine, in-app chat, notifications, user management and audit logging.
• Implemented role-based access control across 12 distinct service roles, spanning SvelteKit route guards, backend endpoint authorisation, permission-aware response shaping and supporting schema migrations.
• Used AI-assisted and agentic development workflows where permitted to compress delivery timelines, holding quality through review standards, typed contracts and test coverage.

02.2HMCTS — Microservice Estate & 3rd Line Support

• 3rd line support across an estate of 70+ services and thousands of repositories; learned Java and Angular on-project and was resolving production incidents in an unfamiliar legacy codebase within three weeks.
• Operate ActiveMQ Artemis messaging infrastructure, including a custom management console built against Jolokia/JMX endpoints; currently delivering Angular components for ongoing service modernisation.

• Researched and prototyped natural-language prompting to extract mixing and mastering preferences, letting users express complex intent without a complex interface, a direct response to user interviews I ran myself.
• Rebuilt and A/B tested the core user interface using Microsoft Clarity session data, lifting conversion 20%; rebuilt key components for mobile after analytics showed 60%+ mobile traffic.
• Launched client-side buffered audio previews so pro users heard results faster.
• Built a deployment pipeline with a Cypress integration suite, cutting critical platform errors by 40% and release cycle time by 25%.
• Resolved escalated customer issues at a 95% user-reported satisfaction rating; owned UI/UX and marketing input alongside engineering.

• Migrated a legacy configuration system to a DAG-based versioned model in Rust, adding concurrent editing and complete validation, which sharply reduced customer issues around promoting config between environments.
• Led full isolation of the largest integration test suite end to end, cutting re-runs from colliding tests by around 30% and running roughly 50% faster through added parallelism.
• Practised TDD and held 80% test coverage; monitored services through Grafana, Loki, Datadog and Retool to sustain 99.9%+ uptime.
• Ran core team ceremonies, including standups, retros, post-incident reviews and support handover.

• Architected and led a from-scratch Akamai EdgeWorkers integration that secured a contract worth over $1,000,000, profiling it to fit hard limits of 2.5MB memory and 20ms CPU.
• Shipped JavaScript edge logic across three CDN platforms (Cloudflare, Akamai and AWS CloudFront) to cut latency and database load; load tested the backend for 100,000+ additional RPS.