Industry
Agri-Tech . Deep Tech
Role
Senior AI Native Product Designer and Strategist
TEAM
Founder-led Concept . Expert advisors in technical feasibility and domain validation
Expertise
Product Vision . System Architecture. Cross-domain translation . Complex data & sensor visualisation . Digital Twin Pipeline
Tech Stack
Figma · Claude Code + MCP servers for AI-first Design. Gemini. Unity Engine
A Farm That Runs Twice. Once in Simulation.
Indoor research farms are the most instrumented growing environments on earth. Thousands of sensors per facility, Robotic fleets collecting data continuously. Every experiment costs a harvest, weeks of grow time, real biomass, and almost no room for a wrong call on light, dosing or robot handling.
Agronomists can't attribute a treatment effect when environmental variance across racks goes unquantified. Robotics engineers can't validate a fleet against a facility that only exists as hardware. Both teams work from the same rooms and different systems.
So I built one model both can stand on. Crop cohorts and robot fleets in a single spatial twin. Physics-based, continuously calibrated, running ahead of reality.
Predictive Digital Twin Dashboard
Agronomists running an indoor farm don't need more data. They need to know what's going wrong, where, and how much time they have to act. The Home screen is built around that question. It brings together the farm's live state, its climate risks, and its crops in one view, so decisions happen before a problem turns into crop loss.
Live 3D Canvas
The centre of Home is a measured twin of the farm: a live 3D model fed by sensors throughout the facility. Agronomists can switch between heat map, airflow and thermal modes to see conditions the eye can't see. A warm pocket behind a rack or a dead zone in circulation becomes visible, and it's tied to where it sits in the room.
Alarm Log
This is the most important part of Home. The Alarm Log shows emerging conditions that could lead to mould before it appears. Agronomists see where the risk is building and can step in while there is still time.
Corm Cohorts & Climate Summary
Each cohort's status sits next to a summary of current climate conditions. This connects what the environment is doing to what the crop is going through, without switching screens.
Together, these give agronomists the essential picture at a glance: what's happening now, what's at risk, and where to act.
"I think once we validate all our cultivation processes, we should definitely get a system like this to steer our climate and practices and make more timely decisions."
— Agronomist, feedback on the concept
AI Did the Production. The Decisions Stayed Human.
Role & responsibilities
As a founder-led conceptualizer I shaped the problem space, the value proposition and the positioning for a predictive digital twin aimed at agronomists and robotics engineers in indoor farming.
Designer in the loop
The AI tooling handled production. The design decisions did not. My role on this project was judgment and verification, proofing every result against the scientists and users who would depend on it. That division is what made the speed safe, and it cut my design cycle roughly in half compared to previous projects where I was drawing the UI myself.
Starting with the people who run the room
Requirements came directly from R&D scientists, agronomists, and robot engineers. Their pain points set the scope: what decisions they make daily, what they can't currently see, and where a wrong call costs the most.
One architecture before any screens
Rather than designing features per role, I built a single UX architecture covering the whole system. How state, forecasts, and interventions connect across dashboards used by very different specialists. A shared model of the farm needs a shared model of the interface.
Mid-fi, tested, then hi-fi
Mid-fidelity designs went in front of users while workflows were still cheap to change. Hi-fidelity followed only once the interaction logic held. The priority was accurate workflow and smooth interaction over visual polish in a tool where a misread number costs a harvest, correctness is the aesthetic.
Code-and-AI-first tooling
The project ran through Claude Code and Figma MCP with Gemini models alongside. Design system guidance . Dev forward Google Material Design, plus a trial of Apple's Liquid Glass language was fed in as structured Markdown rather than maintained as static component files, so the system stayed queryable by the tools building against it.
