I build web, mobile and data-intensive software, combining hands-on engineering with product ownership and technical leadership. Since starting freelance work in 2018, I’ve worked across application architecture, shared services and production operations.
I take ownership of product decisions, working with clients, engineers and business leaders to turn competing requests into a clear plan. I weigh user needs against technical constraints, dependencies and the team’s capacity.
I provide technical direction and work alongside other engineers on implementation. I break down complex work and mentor through pairing and code review, helping colleagues develop their own solutions.
Built and delivered B2B web platforms used by 1,000 monthly active users across 20 businesses, and mobile software supporting 3,500 monthly active users.
Enterprise deliveryLed multiple projects from requirements and system design through delivery and client sign-off.
Backend performanceResolved production bottlenecks, including a 40% reduction in p95 response times for field-data requests.
Interactive mapsBuilt interfaces for analysing tens of thousands of entities, reducing map-loading times by approximately 30%.
Data & machine learningBuilt a reproducible pipeline processing 200 million+ field-level satellite observations for model training and evaluation.
I owned the mobile platform across iOS and Android, from application development and backend integration to identity and production delivery.
The React Native application launched in 2021 after approximately four months of development and later supported 3,500 monthly active users.
Built navigation, core workflows and internationalisation in React Native. Extended Expo into native iOS and Android layers and resolved library integration issues.
Preserved live data and existing login methods through a simultaneous application and backend transition, using Firebase account linking to reconcile previously disconnected identities.
Managed app-store rollout and the production GCP/Firebase setup, reviewing backend changes that affected mobile behaviour and release quality.
I designed and led delivery of workflows that reduced thousands of individual planning actions across a team to a handful of supervisor clicks.
Supervisors could apply a template to a collection, generating schedules and assigning work across its applicable members in one operation. Task lists and Kanban views supported execution, with traceable links back to the source plan.
A season containing crop zones. One template generates activities for each member relative to its own planting date, retaining the shared template source. The three activity offsets are day zero, 14 days and 35 days.
Translated operating needs into user journeys, system requirements and prototypes. I agreed the architecture with technical leadership, led implementation, and used staged client reviews and sign-off to keep scope, delivery and QA aligned.
I designed and built ThreadSense, an independent AI reporting alpha that turns fragmented communications into structured, source-linked findings for investigation and review.
Built with Django and PostgreSQL, it separates ingestion, conversation grouping, model-provider integration and report validation.
Built schema-constrained model-provider integration with timeout, error and usage handling, plus validation of source references and reporting periods.
Added evaluation tooling for usefulness, source grounding and review burden, with configurable retention of source content.
I owned product definition and frontend delivery for a progressive web app combining bookings, machine telemetry and work verification, collaborating with backend engineers on API changes.
In 2024, I delivered the new frontend in approximately two months. We moved application-specific requirements into an existing core API, retired the overlapping API and retained the Google Cloud infrastructure.
Integrated three machinery-data providers through a common representation and shared processing layer, supporting real-time movement and fuel usage without exposing provider formats in the interface.
Used machine tracks to approximate worked area for irregular fields and fields with maintained areas, where parcel area could misstate completed work. Partial completion informed verification and area-based costing.
Later, parcel-entry detection created work activities automatically. Users could arrange bookings elsewhere and still receive tracking and work history without manual activity entry.
I developed satellite monitoring software to help investigate crop-residue burning under Thailand’s no-burn policies.
The work combined NASA fire detections, spatial search and interactive maps with workflow design for field-level review and follow-up.
Used NASA FIRMS data through Google Earth Engine, with clustered views of overall fire activity and drill-down into individual areas, size and intensity.
Integrated signed vector-tile endpoints with spatial and date filters. OpenLayers provided vector-tile rendering and zoom-dependent styling.
Defined how reviewers could link detections to nearby fields, investigate potential breaches and inform follow-up and payment eligibility decisions.
I built a reproducible pipeline processing more than 200 million field-level satellite observations across three growing seasons.
The retrospective MSc research combined satellite imagery, weather and field geometry to forecast field-condition deterioration five days ahead, from acquisition and modelling through reproducible inference and an interactive application.
Two satellite observations, five days apart. The colours show vegetation greenness, measured by NDVI. The third panel shows which areas became affected, stayed affected, recovered or remained unaffected.
Combining satellite and weather data improved probability estimates over weather alone in the overall comparison. It did not show a clear advantage over Sentinel-2 imagery alone. The chart compares prediction error using Brier score: negative values favour the first model, and the lines show 95% uncertainty intervals.
Processed satellite observations with column-oriented storage and reusable pipeline outputs, making large-scale experiments repeatable.
Developed models on 2022–2023 data and tested five satellite and weather configurations on 6.24 million field-date examples from 2024, keeping calibration separate from final evaluation.
Built a full-stack platform for regional-to-parcel exploration, historical planting timelines, interactive filters and shareable maps, supported by spatial APIs and vector tiles.
Built an inference demonstration covering 7,287 fields and a separate inspection-priority workflow using current anomalies, data quality, crop stage and weather.
I diagnosed production failures across mobile applications and shared backend services.
Profiled repeated database queries and other bottlenecks, reducing p95 latency for field-data requests by 40%.
Traced Android failures through incomplete native logs and device/OS differences. Targeted fixes reduced the crash-affected session rate by 10%.
Owned mobile release automation and Firebase configuration; co-developed reproducible cloud delivery with GitLab CI/CD, Docker and encrypted configuration. Automated builds and deployment replaced a day-long process with an approximately one-hour pipeline.
Introduced Agile/Kanban planning and shared release documentation in GitLab, and helped secure dedicated part-time QA support.
I contributed to shared Rails services backed by PostgreSQL and PostGIS.
Contributed to organisation-scoped APIs and data models, authenticated WebSockets and viewport-driven vector-tile endpoints.
Contributed to satellite ingestion across 100,000 fields: historical backfills, quality checks and Pub/Sub workers with transactional updates and completion acknowledgements.
Co-developed activity suggestions using similarity search, organisation-specific context and validated structured responses.
I led development of map interfaces for analysing tens of thousands of entities across five analytical layers, shortening map-loading times by approximately 30%.
Custom OpenLayers WebGL loading and per-image state coordinated imagery swaps, reduced flicker and prevented redundant layers.
Combined caching, debouncing, batching and spatial clustering to control data loading across zoom levels.
Extended the soil-analysis interface and state model from one sample per parcel to multiple samples.
I direct AI-augmented investigation, planning, implementation and review, retaining responsibility for architecture, integration and acceptance.
I give specialist agents bounded tasks, relevant context and clear ownership, coordinating parallel work through worktrees and concise handoffs.
I separate implementation from review and assess changes against requirements, tests and source evidence, managing context and token budgets with explicit retry limits.
Web, mobile and shared-platform software for agricultural businesses and enterprise clients.
Delivered customer-facing React components and modernised payment integration.
Website and internal database improvements informed by customer feedback.
Built web features, including scheduling across time zones, and improved API performance.
Taught web development, coached projects and reviewed interfaces.
Teaching, campus coordination, curriculum discussions, colleague development and parent communication.
University of Bath
Le Wagon
University of Nottingham
Loughborough University
Interfaces & applicationsReact · TypeScript · Next.js · React Native · Expo
Systems & dataRails · Python · PostgreSQL / PostGIS · APIs · WebSockets · DuckDB · Parquet
Applied AIData pipelines · model evaluation · probability calibration · retrieval · structured generation
Delivery & diagnosisGCP · Firebase · GitLab CI/CD · Docker · Nix · New Relic · Rollbar
FROSTLINE / OFF THE CLOCKSend a message or request my CV.
hello@jupiterslb.comContact: jupiterslb.com/#contact
The island artwork couldn’t load.
