Sprout

Dashboard

Inspiration

Roughly 50,000–60,000 companies incorporate in Singapore every year, and a large share of that flow starts the same way: a founder in Jakarta, Ho Chi Minh City, or Manila who wants to expand into Singapore but has to reverse-engineer the answer to several questions like structure, visa, licensing, tax, banking, timeline and risk. Given from the scattered information on Singapore government websites it is hard to piece everything together because it all boils down to the niche on what you company have.

What it does

A founder answers five short questions (home country, industry, entity purpose, founders/staff relocating, projected Singapore revenue, EntrePass evidence) and gets back a full playbook, which consists of entity structure, licensing, tax, banking, timeline, and risk matrix.

How we built it

We split the product into two halves that never mix: the actual regulatory answers are computed instantly by rules we wrote ourselves based on real government sources, so those numbers are never something an AI is guessing even with no internet connection. On top of that, we use an AI model to write in words explanation of those numbers, personalized to the founder's specific situation, but it's only ever allowed to describe facts we've already calculated

Challenges we ran into

The single biggest challenge was almost shipping a demo that was factually backwards. Our first pass at the COMPASS benchmark had a startup scoring 0 on C3/C4 and failing at 20 points which essentially meant a clean, dramatic story. It was wrong. MOM's actual small-firm rule means that same startup scores a neutral 40 and passes. This also mean re-verifying the rule against MOM's own pages and rebuilding the scorer around it rather than revolving around the more wrong answer.

Accomplishments that we're proud of

Catching and correctly encoding the small-firm COMPASS rule before it is deployed
An architecture where can the AI hallucinate a regulation? has a real structural answer.
Three demo profiles that each tell a genuinely different product story instead of three shallow variations on one.

What we learned

For compliance AI, rules engine computes, LLM narrates is a much stronger answer than LLM generates everything and we hope it's accurate. It's also a better answer to why do you even need AI here?

What's next for Sprout

Expand the knowledge base beyond the five demo verticals (F&B, SaaS, retail, medical devices, fintech) toward the full industry table.
Add more mock profiles or a bring your own profile mode that isn't gated behind the intake template.
Explore adjacent monetization the market size, for instance, corp-sec firm and bank partnerships, EDB collaboration for verified incentive matching.