"Boring stuff"
Work

Twelve weeks, one pivot

Brought in to do applied AI research. Ended up on a key production module for a cross-agency project.

What I did

This internship was part of my sponsorship contract with IMDA.

I was initially brought in to do legal tech, looking into neurosymbolic AI for a legal verifier. I read papers to validate my approach and had a presentation ready to go. Then I was moved to a different, more urgent production project with looming deadlines.

We worked on different parts of the pipeline. I built a supporting harness: a knowledge base that adds human expertise to the instructions generated upstream. Those instructions then fed downstream, into my friend’s module, where our LLMs generated 3D models.

I experimented in phases, from hand curation to learning from human-made models. There was a lot of iteration. Not every trial worked, so I read papers to reason about why some things worked and others didn’t. And of course, I logged it all and explained it in a way my supervisor’s stakeholders could understand.

It was an interesting 12 weeks. It’s the closest I’ve come to “research” work: reading papers for inspiration, running experiments, building proofs of concept, validating, then iterating.

What I learned

  1. Presenting an idea is as important as doing it. The three of us (my friend, the full-time AI Scientist and I) worked on this project full-time with my supervisor, the AI lead, so we knew what was going on. Not everyone could follow, though. When we presented to our product owners, the main feedback was that “it was too technical” and they “couldn’t understand what each specific module did”. (Not that they needed to know that anyway.) I learnt that explaining ideas to different kinds of people is a skill, and a very important one.
  2. AI moves ridiculously fast. GPT-6 Astra came out mid-project and blew everything we had out of the water. Two months of my work was (kind of?) replaced. But, I’m glad I could adapt and give my harness a new life in a different form.
  3. Open eyes, open mind. I didn’t want to move projects at first. Accepting it and coming at the problem with an open mind taught me a lot about harnesses, and inspired me to build my own.