AI Engineering Intern

INNOVATION: AI Engineering Intern

Own the evaluation layer of Sherpa, our AI business validation tool, used by founders and students across Asia Pacific.

LOCATION: Remote (Singapore or Johor Bahru)

DURATION: 3 to 6 months (Min. 3 days a week)

START: Immediate (Paid)
REPORTS TO: Head of Engineering, InnovNation.ai
WHY THIS ROLE EXISTS

InnovNation.ai is an innovation and entrepreneurship platform working with universities, incubators and early stage founders across Asia Pacific. Sherpa, our AI powered business validation tool, challenges founders on their assumptions and generates business model canvases from their inputs. We also run a course builder for partner educators and community spaces for campuses in Malaysia, India and the Philippines.

Sherpa is used by people who make decisions with it. That raises the bar on two things most AI products treat casually: knowing whether a change actually improved output quality, and knowing where the model's competence ends. This role owns both.

WHAT YOU WILL WORK ON

Evaluation: Build the eval set that lets us tell whether a change made Sherpa better or just different. Define what "better" means for canvas generation and for challenge questions, build the test cases, and turn them into something we can run repeatedly rather than judge by feel.

Sherpa's output quality: Improve prompts and output structure, especially canvas generation and the challenge question flow. Every change goes through the eval set you build.

Adversarial testing: Run Sherpa against realistic founder inputs, including incomplete, contradictory and vague submissions. Document where it fails, in enough detail that someone can act on it.

Local and regional accuracy: General purpose models reason about our region badly, and usually with total confidence. They default to Stripe and ACH where the answer is PayNow, DuitNow or GrabPay. They reach for GDPR where Singapore's or Malaysia's PDPA applies. They flatten MAS and Bank Negara into one imagined regulator, and they do not know what Singpass, Myinfo or e-invoicing under LHDN change about how a startup actually operates. The same class of failure repeats in the Philippines and in India, where the relevant frameworks are different again. Finding these gaps and closing them is real, valuable work.

Course builder support: Text to speech and script generation for our partner educators.
Writing it up: Findings that non engineers on the team can read and act on. Clarity counts as much as the analysis.

WHAT WE ARE LOOKING FOR

• Hands on experience with LLM APIs, prompting and structured outputs
• Python or TypeScript
• Real interest in evaluation and measurement, not just prompt tinkering
• Intellectual honesty about model limitations (we would rather Sherpa admit a gap than invent a confident answer)
• Ability to write clearly for a non technical reader
• Bonus, not required: exposure to RAG or agent frameworks, or any market, policy or regulatory research background. Familiarity with the Singapore or Malaysian startup, fintech or edtech landscape is a genuine advantage.

WHAT YOU DO NOT NEED

• A computer science degree
• Prior formal internship experience
• Published work or a large GitHub following
• Existing knowledge of Indian or Philippine markets (you will learn those on the job).

WHERE YOU CAN BE BASED

Remote, with a few hours in person each week. In practice that means Singapore or Johor Bahru. If you are elsewhere in Malaysia and can genuinely commit to being in Singapore weekly, tell us in your application and we will consider it.

WHAT YOU GET

• A paid internship of three to six months, minimum three days a week, starting immediately
• Direct work with our Head of Engineering, with weekly review and real ownership of the eval layer
• A portfolio piece that is genuinely rare: an evaluation framework you designed for a production AI product
• A path to a full time role for strong performers

HOW TO APPLY

No cover letter. Instead, a short exercise, and it should take under an hour.
Pick any LLM you have access to. Prompt it with a deliberately vague startup idea, for example: "an app for helping small shops in my area." Then send us one page covering:

1. Five test cases you would use to determine whether a change to that prompt made the output better rather than merely different
2. What each test case is actually measuring
3. One thing your test set would fail to catch

Email it with your CV to [email protected], subject line AI Engineering Intern.