Young people will have new good ideas
I just ended my IMDA internship a few weeks ago, and I thought it’d be nice to write about it, before I forget the smallest of things that made my day, like iJOOZ (orange juice).
Part 1: How it all began
Funnily enough, I actually met my supervisor at a grants pitch. Fun!
My friends run a legal-tech startup, and they wanted funding from IMDA. They knew IMDA sponsored me, so I became the token “Cambridge CS Engineer” on the team. I guess they wanted to improve their odds of getting the grant. (I think they got it?)
Anyway.
After that presentation, I asked my supervisor to bring me in, since I had to intern with IMDA as part of my sponsorship contract, and it would be great to get some experience in “AI Scientist” work (which was his title). Thankfully, he did! That’s how I ended up in IMDA’s BizTech Group.
I was initially brought in to do legal tech, looking into neurosymbolic AI for a legal verifier. This was right up my alley. I was fresh out of a couple of legal-tech hackathons (with the same friends), and I had seen projects that did similar things.
I started reading papers to validate my approach, and I had a presentation done up and ready to go.
Part 2: “Young people will have new good ideas!”
Then I was moved to a different, more urgent production project with looming deadlines. My supervisor’s reasoning, and I quote: “I think you need to help him [the other full-time AI Scientist]. Young people will have new good ideas!”
So my friend and I were placed on this project. (Yes, my friend also kenna. Kenna is Singlish for “having to go through something”. The translation isn’t perfect, since kenna usually has a more playful, negative ring to it. But yes, where were we?)
I was quite unwilling to move at first. (Give me a chance, man. I did all the prep work and was about to start the exciting stuff.) But it’s about picking yourself up and moving fast (and failing fast).
Part 3: Outpaced
My part of the project was a knowledge base. It was our take on a harness: the tools, agents and so on that support the model. It was meant to support models like GPT-5.6, which had weaker spatial reasoning. And it was taking shape.
Then GPT-6 Astra came out, and my friend ran some trials with it. Astra blew everything we had out of the water. It outperformed even humans in benchmarks on spatial reasoning, and that’s my hypothesis on why Astra is (comparatively) so much better at 3D modelling than its predecessors. So the knowledge base had to take a different form. It went into plugging the gaps Astra left, and was redesigned to be minimal.
That meant two months of my work was (kind of?) replaced, which felt awful.
But it showed how fast the AI scene moves, and I’m glad I could adapt and give my harness a new life in a different form. Accepting the situation and coming at the problem with an open mind taught me quite a bit, in particular about harnesses and keeping up with AI model development. It made me more conscious of skills, subagents and tool use, and it even inspired me to build harnesses for my own work! (Going to try building one with Pi.)
(the nice) Part(s)
No numbering needed here since this isn’t really about the work I did. Please appreciate the attempt at a consistent format though.
There were so many nice things in IMDA, I don’t even know how to start. In no particular order:
-
iJOOZ was great :> We were entitled to one orange juice a day (if I turned up in office), and it was nice to have “OJ” (as we called them) runs with the other interns. Kind of like a daily pilgrimage (when we were all in office), which made it extra nice!
-
Welfare events! There was “Thirsty Thursdays” in July, the Durian Fiesta in August, a movie screening in September (and a very good movie at that: The Violinist is the title). These were very nice indeed.



- Definitely the people. My intern friends, my supervisors (both my actual one and my honorary one!), my team, and just everyone in general. They made the internship
more bearable I meangreat!


Part final
So what did I learn? Quite a bunch: I refined my way of using AI, I learnt about how sprints function, I learnt about how presenting (and communicating) is as important a skill as being technically competent.
Maybe it is true that young people do have great ideas! All we need is a little more belief in ourselves, and more controlled, thoughtful use of AI.