GEMINO AI
AI Research Intern — Seraph Research Cluster, EX Venture — Canggu, Bali
Get notified of more jobs like this · No spam, ever
How your CV stacks up
Upload your CV to see how well it fits this job role
?%
Seraph Research Cluster · EX Venture · On-site, Canggu, Bali · 4–6 months
What does it take to make a network of on-device agents reason together without a server in the loop — and catch the exact step where one agent hands another a conclusion it can't justify? This is an AI Research Intern building on-device multi-agent reasoning for the Seraph research cluster — the funded multi-partner AI research cluster, UTN Nürnberg among the partners, at EX Venture. EX Venture is the operating company; Epic Solutions PT runs the on-site program. Most multi-agent research stops at the cloud, where logs are clean and latency hides the hard failures. The job is to take the reasoning onto devices and keep it honest. You leave with real execution experience — a thing you made that nobody asked you to make. Everyone has a degree. Almost nobody has proof.
What you build
Not a task list. The thing you build.
- You build a local multi-agent reasoning harness that runs several agents on one device and logs every handoff.
- You work on a routing protocol that decides when an agent can resolve locally and when the chain must pause and ask for a stronger model.
- A replay pack turns a real reasoning failure into a reproducible benchmark the next cohort can rerun.
- Reading the logs shows which agent carried the correct answer and which one introduced the drift.
Reasons to use Rodeo
Start with a chat, not a search bar
Grad scheme, placement, apprenticeship? Not sure what you want yet — that's fine. Your agent talks it through with you and turns "I have no idea" into a shortlist.
It searches the market for you
Every day your agent scans the market matching roles against what actually matters to you, not just keywords on a CV.
Only hits
No noise. No "maybe this fits." Just roles with a clear explanation of why they're right — and where to focus when applying.
The bar
- You have shipped an agent or a reasoning loop outside coursework and can name the exact step where it breaks.
- A harness another person can run without your notes.
- A benchmark that turns one real failure into a reproducible case.
- Kept a hard reasoning problem on the table until the wrong answer could be traced to one faulty handoff.
Who this is for
Engineers and researchers who treat on-device multi-agent reasoning as a system problem, not a paper title.
- You run the harness until the wrong answer can be traced to one faulty handoff.
- A failed run is data, not judgment.
- The work continues until the reasoning is defensible.
Where you do it — and why here
Thirty years building companies — New York, London, Tokyo, the Valley — and then it stopped mattering where the work happened. So the program runs on-site in Canggu, Bali. The campus is 4,000 sqm, and the people who run the companies are on it every day. A modern company costs twelve people and a toolchain — so being wrong six times a year is part of how we build. Monday reset, Friday ship: send what you built. You know Bali as where you switch off; here it earns your full attention. One cohort, on-site, the full phase, not remote. What this location buys is attention, not a view.
What happens after
Seventeen interns have moved into full-time roles across the group. This research cluster is phase one — we are early, and we say so. The destination role and its terms (role, visa, pay) are agreed before you commit, never after. The internship is step one, and paid work is the destination for the ones who stay.
Get help with your application
Your very own career expert that helps elevate your application to the next level.
Terms
Unpaid under Indonesian law. A room is available on campus — rent paid by you. Epic Solutions PT hosts the on-site program and assists with the C22B internship visa process; the visa fee is yours. Daily lunch on us. Starting October.
Questions about the fit or the timeline? Tell us — we'll talk you through how the program actually runs.
Requirements
- A degree in computer science, engineering, or equivalent — you have proven you can finish something.
- You drive Claude Code, Hermes, or Codex every day and catch what the AI gets wrong before the harness ships.
- Not what you know, but how fast you learn — the reasoning stack changes monthly and you stay ahead of it.
- A thing you made that nobody asked you to make: an agent, a benchmark, a harness, a run log.
- Four to six months on-site, the full phase.
How to apply
Apply via the LinkedIn application field. Submit your CV and the repository or benchmark you are proudest of.
A finished CV was never the point. What counts is a thing you've built and the drive to build the next one. That is who we are looking for.
“It took my CV and asked me questions relevant to understanding what kind of jobs to suggest for me. Suggestions were almost perfect. Jobs were exactly what I’ve been looking for.”
Jessica, London
Location