Started with Impact Collab's Expert Pod in late 2025, eight impact-led founders coached through one-week pods, four detailed proposals delivered. Converted to paying engagements in early 2026; seven paying clients so far.

Since September 2026 the software is free and open source. What I offer is help setting it up: a walkthrough (an hour, then 30 minutes) or a guided setup (an hour, then two hours), each with 30 days of Slack support. You leave owning the system and knowing how to solve problems with it.

Architected an end-to-end data pipeline to lower the barrier to entry for environmental impact assessments. The team's messy multi-source data lake became a three-tier warehouse, raw, processed, analytics-ready, with scientist-facing monitoring interfaces.

Delivered using Claude Code as an AI development collaborator. The approach was systems-thinking plus rapid iterative delivery in a team used to constrained, precision-first scientific work. The science team's domain knowledge was facilitated into operational tooling rather than replaced.

One public component: an 8-step river mesh-generation pipeline at github.com/cradsdavis-cell/mesh_generator.

Python PostgreSQL Claude Code Geospatial

Three years tutoring second-year engineering students in mechanics and dynamics, small-group teaching, exam-marking, course assistance.

Known for an engaging, intuitive style and psychological safety in the room. Asked by the course coordinator to author the full second-year exam.

Three-and-a-half years at the DARE ARC Training Centre developing BayeSpace, a Bayesian statistical framework applied across atmospheric dispersion modelling and species distribution modelling.

Two novel empirical findings landed in the thesis: oceanic atmospheric dispersion appears more intense than Pasquill–Gifford terrestrial models predict; and a 20% spatiotemporal grid-coverage threshold below which species distribution data becomes unusable.

Presented at six conferences. Industry placement at Xylo Systems (biodiversity AI startup).

One-week sprint with the Centre for Environment, Fisheries and Aquaculture Science (Cefas). Facilitated a team of ten data scientists from scratch to a working pipeline by Friday, with deliverables across feature engineering, model selection, and a final presentation to the Cefas marine team.

The sprint shape stuck. The same pattern shows up in the AI consulting work and in the Wildly Calm retreats: take a group through something hard, hold the room, and ship by the deadline.

Six-month placement applying a genetic algorithm to optimise the geometry of a small satellite antenna array. Built the optimiser and a UI to drive it.

Antenna optics was new, taught it to myself in the first month from journal papers and textbooks. The same self-teaching shape repeats across the career: orbital mechanics at ESA, Bayesian methods at DARE, Claude Code and MCPs in 2025–26.

Nine months at a Manchester mobile-software consultancy after the MPhys, production software engineering across multiple languages and frameworks. Foundation for the end-to-end data and AI systems work that followed.

Left to take the PhD route at Sydney via the DARE ARC Training Centre.

At 22, a summer intern at ESTEC working on the FLEX Earth Explorer mission. Calculated the delta-V budget and propellant requirements for the satellite's operational lifetime.

Built a probabilistic Monte-Carlo alternative to the standard worst-case method. Invited by the team to co-author the approach as a scientific paper. Orbital mechanics was new, taught it to myself from textbooks in the opening weeks.

A year between school and university teaching A-Level physics and maths to Year 12 and 13 students at Harrow International Bangkok. Pastoral duties in a boarding house. Ran a weekly rock-climbing programme.

Organised the school's annual charity event (A4D). First serious teaching experience, the draw was the holistic shaping of young people, not the subject delivery.

The thesis developed BayeSpace, a Bayesian statistical framework applied across atmospheric dispersion modelling and species distribution modelling. Two novel empirical findings: oceanic atmospheric dispersion appears more intense than Pasquill–Gifford terrestrial models predict; and a 20% spatiotemporal grid-coverage threshold below which species distribution data becomes unusable.

Trained as part of the DARE ARC cohort, the federal training centre for data science in environment-and-water research. Six conferences including the Alan Turing Institute Data Study Group. Industry placement at Xylo Systems (biodiversity AI startup).

Research ran 2022 to 2025; the degree was awarded in 2026.

Joint degree in Mathematics and Physics, awarded First Class. Placed 4th in the physics cohort at the end of third year (cohort of ~250–300 at A*A*A entry).

Two years at the Welsh-medium sixth-form college in Dolgellau, north Wales. A* in Mathematics, A* in Further Mathematics, B in Physics; AS-Levels in Geography, Chemistry and the Welsh Baccalaureate.

A Welsh-medium state comprehensive in Blaenau Ffestiniog, in the heart of Eryri (Snowdonia) in Gwynedd. Five years of secondary schooling through GCSE.

Non-elite-to-elite contrast that recurs across the career: state Welsh-medium → Harrow Bangkok at 18 → Manchester MPhys → PhD at Sydney.

The assistant I set up for clients and run my own week on. It lives on the owner's own server or computer, under their own accounts, and connects to 63 services, Gmail, Google Calendar and Slack among them.

Free to download, and every line is public at github.com/cradsdavis-cell/crads-ai.

Claude Open source Desktop app

A private wiki of everything I know and decide, wired into my email, calendar, tasks, messages and meeting notes. Every morning it tells me what's slipping. It drafts in my voice and keeps track of every client.

It runs my client work, my job applications and my planning. Everything I set up for clients was tried on my own week first.

Claude Code Obsidian

A carbon-accounting prototype for the UK consultancy Greener Edge. Eleven specialised agents handle scope-by-scope intake, factor lookup against Climatiq, methodology selection, audit logging and the customer-facing report.

The most sophisticated of the 2025 builds. Never productised, Greener Edge re-engaged on a different shape in 2026.

PostgreSQL Climatiq API RAG Multi-agent

A conversational research assistant for biodiversity questions, pulling structured occurrence data from GBIF and synthesising it against Perplexity-sourced literature. Built in n8n.

n8n GBIF Perplexity OpenAI

One public piece of the SEAF data work, at github.com/cradsdavis-cell/mesh_generator.

Python Geospatial

Co-founded with Lockie Ranson and Kieran Maguire in 2025. Retreats since then include Camp Bunya, Malabar Leaning Circle, Croajingalong and the Snowies, plus ongoing community events. The retreats, past and next, are at wildlycalm.org.

The format proven at Croajingalong: take a group through something hard, build a safe container inside the discomfort, expand and learn within it, integrate on the way back. Branding led by Rosie.