Rob Gilks

• Led the TAPI team within Assessment & Research Capabilities (ARC), building and operating the Text and Speech APIs that power automated marking across Cambridge English assessments, including Write & Improve and Speak & Improve - and stayed the largest individual contributor to the Text API while leading it. The platform held 99.98% uptime across twelve months at around a 1% change-failure rate.
• Operated the response-validity suite that decides whether a submission can be marked at all - off-topic, gibberish, insufficient-English, offensive and repetition detection, plus copy-detection. The detectors were the ML teams' work; running them dependably in production was mine.
• Instigated and built two of the platform's systems: Text Orchestrator, the Clojure/AWS service that drives automated text assessment, and the multi-account automarking administration dashboard in Next.js, React and TypeScript on AWS Amplify/Cognito.
• Productionised Python NLP/ML inference as Dockerised SQS workers with CI/CD, diagnostics, log redaction and LocalStack/AWS tests, hardened the services around them with retries, idempotency and alarms, and was one of the most active contributors to the Terraform monorepo underneath.
• Owned my team's part of the Cyber Essentials Plus compliance programme - the assessment APIs - covering AWS account scope reduction, CVE and IAM remediation, secrets, backup and disaster recovery, and a written security incident response procedure.
• Moved the team towards continuous delivery - deployment separated from release behind feature flags, four environments gated by automated acceptance tests, and DORA metrics tracked with a tool I wrote and open-sourced. Ran an Accelerate reading group to build the case.
• Built the team's documentation culture - architecture and diagrams as code, RFCs, runbooks, incident response and a public API documentation site - and introduced its AI-assisted code review process, with automated pre-review on every pull request and risk-tiered approvals so human review concentrates on intent and operational impact.

My own company since 1998; full-time on it since leaving Cambridge.

• Building and running live AI products for language learning and assessment (below) on Cloudflare Workers, with serverless GPU inference on Modal.
• Fine-tuning and deploying transformer models on serverless GPUs, with post-hoc calibration and grammar correction alongside the scorer.
• Choose and gate models on measured evidence: frozen benchmarks with held-out folds, fail-closed quality checks on generated content, and side-by-side model comparisons scored on task success, convergence and failure modes.
• Build throwaway projects to learn things properly. The product is often trivial; the point is having a real reason to work out how something behaves.

• Senior engineer on Speak & Improve - an AI feedback tool that helps English learners worldwide improve their speaking, built in Clojure around custom marking models.
• Moved to the API team, which became TAPI within ARC following ELiT's integration into Cambridge in 2025.

• Led four to six engineers while remaining hands-on and owning delivery for a system that enabled clinical research sites to communicate with trial patients by SMS.
• The role ended when OneStudyTeam cut 40% of its staff in a single day, citing changes in market conditions.

Concurrent contract roles for multiple clients through my company.

• CTO, Limpid Markets (2018-2020) - set the technical direction for a precious-metals derivatives trading platform and transitioned software delivery from a London team to a Ukraine-based team.
• Technical Lead, Predira (2020-2022) - built the early prototypes of a customisable brokerage trading platform with the founder, then grew the team and engineering practices.
• Technical Lead, SimPlay (2019-2021) - rebuilt a platform for running collaborative role-play training experiences online.

• Ten years building interactive products and campaigns for Nike, Heineken, Nissan, the BBC and Ford - including Heineken Star Player (Cannes Gold Cyber Lion) and the BBC's Story of Life.

• LTA - TotalTennis club website platform and live tournament scoring (Nottingham, Eastbourne). Oxford University LTG - Java learning tool middleware. Play Sport / Pretzel Logic - sports club and government web systems. Ampac Australia - fire-detection systems, from electronics technician to R&D engineer.