Ewan Trollip
I’m based in South Africa and lead development on SurveyScope at WildEye Conservation. Lately, I’ve been spending a lot of time getting AI agents to review, test, and fix their own code.
SurveyScope
This is the project I spend most of my time on. I lead its development at WildEye Conservation, where we’re building software to take aerial wildlife surveys from photographs through to population estimates.
A lot of photographs to work through
Each flight leaves us with thousands of images. Model detections give reviewers something to start with, rather than having to annotate everything from scratch.
The models miss things
Part of the work is giving reviewers a way to find those misses, correct annotations, and check the results.
Making sure it’s the same elephant
An animal can appear in more than one photograph. SurveyScope helps align the images and link those sightings before the counts are used to estimate the population.
EFT Toolset
I kept looking things up while playing Tarkov, so I built a site for it. Standing still at any point in Tarkov literally puts a target on your head. If you’re gonna look things up, it should be blazingly fast. I give you EFT Toolset.
The things I keep needing to look up
What an item is for, which quest needs it, and what the traders have available.
Built for my phone
I built it with Next.js, TypeScript, Tailwind, and shadcn/ui. I use it on my phone while playing.
Teaching & Learning System
A group project at North-West University. We built a system where students could submit video assignments and lecturers could review them and leave feedback.
Web, mobile, and a Django backend
The project was split across three apps. Together, they handled accounts, video uploads, tagging, and feedback.
The lecturer’s side
Once a student submitted a recording, their lecturer could watch it and leave feedback in the same system.
An unreasonable amount of time on agents
My setup is mainly T3 Code with a variety of AI subscriptions. I’ve been adding review and testing tools to see how much the agents can handle before they need me involved.
Sending it back for another pass
The agents can ask for a review, fix what comes back, and try again. Getting that process working has become a project of its own.
The home server is involved now
It runs review and testing services alongside local models, Hermes Agent, and bots. Some of that testing happens in disposable Docker or QEMU environments.
Still figuring out how far this goes
AI has had me working well beyond my usual stack. I’ve never been this motivated to mess with how I work.
Ewan Trollip
- Lead Developer, SurveyScope
- WildEye ConservationJune 2026 – now
- Software Developer
- Innoventix ConsultingNov 2024 – May 2026
- Freelance Web Developer
- Shopify & custom stacks2024
- BSc IT, cum laude
- North-West University2022 – 2024
- Matric, five distinctions
- Noordheuwel High SchoolMatric
Mostly working with
TypeScript · React · Vite · TanStack · Python · AWS · Amplify Gen 2
What’s running at home
T3 Code · Review & test loops · Docker · QEMU · Local models · Hermes Agent · Tailscale
Other tools I’ve worked with
Next.js · Three.js · Convex · PostgreSQL · Supabase · Tailwind · shadcn/ui · Expo · Shopify
University coursework
C# · Java · C++ · SQL Server · ASP.NET · WinForms
Based in
South Africa
Get in touch
If you’re also spending too much time tinkering with agents and automation, we’ll probably have something to talk about.
ewantrollip@webdevewan.com© 2026 Ewan Trollip