My journey

• Building a centralized, scalable platform that brings together AI capabilities, application layers, and infrastructure to significantly reduce the time, cost, and effort required to deploy the solution at new museums and attractions.
• Designing automation-first deployment pipelines and standardized workflows to enable rapid, repeatable rollouts of the point-and-discover experience across multiple cultural venues.
• Actively developing a centralized Aggregation & Observability Platform to monitor all deployed museums from a single portal, unifying visitor engagement, behavioural insights, feedback, and system health.
• Designed and built the self-service museum kiosk (Android 14, Kotlin + Jetpack Compose) that sells on-site access to the digital guide end to end: attract loop, card tap, and a printed receipt carrying the redeem QR, opening a second sales channel that runs without staff involvement.
• Integrated unattended payment hardware and peripherals directly, driving the myPOS Sigma Unattended SDK alongside an ESC/POS receipt printer, QR referral scanning, and NFC admin access, behind a terminal abstraction so payment providers can be swapped without touching the flow.
• Built the fleet management layer for kiosks in the field, covering heartbeat and command polling, remote reboot/refresh/reprint, silent over-the-air updates with staged rollout via device-owner PackageInstaller, a watchdog, and durable on-device logging for crash triage.
• Made the kiosk resilient and locale-aware for real venue conditions: offline ticketing fallback when the network drops, on-device evaluation of museum opening hours so it closes itself outside them, per-kiosk configuration overrides, and simultaneous bilingual screens (e.g. Welsh alongside English).
• Kept visitor presence detection privacy-preserving by processing the camera feed entirely on-device and transmitting only abstract presence events, never imagery.
• Working hands-on with museum visitors on-site, observing real usage of the mobile point-and-discover web app to identify friction, bottlenecks, and usability gaps and drive true product–market fit.
• Defining the platform and product strategy with a strong focus on user-centric design, operational scalability, and continuous learning, ensuring the experience improves with every new deployment.

InnovateUK-funded tech startup building a visitor-centric point-and-discover experience for museums and attractions.

• Fine-tuned and integrated the provided computer vision components (embeddings, ONNX inference, model outputs) into the mobile web application, optimising performance and reducing battery drain on low-end devices by over 50%.
• Designed and built the full mobile-first, edge-optimised web application for a point-and-discover experience used in museums and cultural spaces, including the frontend (Next.js) and backend (FastAPI) along with all supporting infrastructure.
• Developed the complete backend architecture using Python (FastAPI) and MongoDB, covering APIs, data models, content pipelines, and embedding management workflows.
• Built the comprehensive Admin & Content Management Portal, enabling curators and internal teams to manage collections, metadata, assets, geometries, and discovery logic through a streamlined interface.
• Architected and implemented the full CI/CD and DevOps ecosystem using GitHub Actions, with automated builds, deployments, testing pipelines, and environment management.
• Delivered a reliable and scalable cloud infrastructure on AWS, including Amplify, CloudFront, S3, EC2, CloudWatch, and AWS Backup, ensuring performance, monitoring, and resiliency across the entire product.
• Designed and built a rule-based Retrieval-Augmented Generation (RAG) Prompt Engine: a dynamic prompt construction system using definitions, templates, rules, and composable components, powering the conversational AI layer of the application.
• Delivered all core components (frontend, backend, admin portal, DevOps pipelines, AWS infrastructure, integrations, and performance optimisations) within a rapid 4 to 6 month timeline to achieve a production-ready MVP.