Models trained on your data, evaluated, and put into production.
Charles Jackson
AI engineer
I develop machine learning models and automation, and the software around them: websites, web apps and native apps for Mac, iOS and Android. I’m interested in robotics, and in ideas that will shape the future.
How a project runs
What decision or task is being automated, who relies on it, and the number that will say it works.
Where it comes from, how clean it is, what is missing, and what must never leak into training.
The simplest method that could work, measured first, so every later gain is a real one.
Features, architectures and prompts tried against the same test set, with every run tracked.
Held-out data, edge cases, cost and latency. The errors are read one by one.
Pipelines, APIs, interfaces and automation, under test, so the model does its work inside a product.
Released behind checks, watched for drift, retrained when the data moves, and documented for the team that owns it.
What I build

Agents and pipelines that take repetitive work off a team and run without supervision.
Public sites and the applications behind them, front end and back end.
Apps built for each platform, from the interface to the release.
Project
insilico.systems
A robot arm, from the idea to the machine
insilico is robotics research and a native engineering studio. This is its desk arm, drawn in its own CAD.
A hand with five fingers
Every finger has its own joints. The ones moving now are driven from the project’s own motion data.
Taken apart, part by part
Covers come off to the sides, the shoulder and elbow drives open along their axes, and the forearm and hand lift clear.
More than 2,300 parts, all modelled
The hardware is still research and development. The tools that design it are built to last.
The hand, on its own
Each finger comes apart into the links and joints that move it, then goes back together.
- Role
- Design and development
- Built with
- Rust
Project
microhabitat.ai
Private evaluations for crop AI
MicroHabitat tests a team’s model on crop growth, plant health and harvest timing, and reports where it holds and where it misses.
Ten years of field data
Photographs and growing records from urban farms, collected since 2016. The raw data stays in-house; a team receives the findings.
One planter, visit after visit
Each record ties what was observed to the care that followed and to what had changed by the next visit.
The forecast against what grew
A model is scored on visits it has never seen, so the gap between its prediction and the plant can be measured.
- Role
- Design and development
- Built with
- Next.js and PyTorch
Project
cursuum.com
A social calendar for the Mac
Cursuum finds a time, an activity and a place that suit everyone’s schedule and preferences.
Their time, and what they like
Each person shares when they are free and what they enjoy: good espresso, pancakes, something vegetarian, a table outside, somewhere near the park.
One morning all three have
Set side by side, the three months have a single day in common.
A plan that suits everyone
Friday the 13th, ten to eleven, at a café terrace by the park. The people and the café are a sample.
- Role
- Design and development
- Built with
- Electron, Next.js and Supabase