Build an AI Health-Info Helper That Knows What to Refuse
Maps to: AI Application Builder · Physician · Public Health · Health Policy · Medical Researcher · Health Journalist
You're going to build an AI helper that explains health information from trusted places like the CDC and NIH, and that knows when to say no: it never tells anyone what's wrong with them, never tells them what to do, and sends anything serious straight to a real doctor. The skill is deciding what a tool like this must never do, then proving it holds when someone asks the dangerous questions. As these get used for health, that's the rarest thing anyone building one can do.
You're done when: A helper anyone can use that, on every dangerous question (emergency symptoms, someone saying they might hurt themselves, someone asking what illness they have), says no safely and sends them to a real doctor, with a warning built in, plus one page on what you couldn't do safely. Not 'it answers health questions,' but 'it says no to the dangerous ones, and I can show that it does.'
How this shows up on a resume or college app
I built an AI assistant for scope that handles edge cases like emergency symptoms and requests for diagnosis by refusing and routing users to real clinicians, designing the safety eval and proving it fails safe. I learned that the hardest engineering problem in medical AI is what the tool should refuse to do.
When you finish, BuildMe drafts your Common App activity description from what you actually built.
Not sure yet? Play 5 minutes as a ai application builder first and see how the work feels.
The plan
- 1
Step 1
Pick one narrow job, and write the rules first
Pick one narrow health question your helper answers, then, before anything else, write the rules: the warning it always shows, and exactly which questions make it stop and say 'go see a real doctor' or 'call emergency services'. This step is not optional and it comes first. Everything else is built on it.
- 2
Step 2
Write its instructions, then throw 10 hard questions at it
Write the instructions that tell it how to behave, pointing it only at trusted places (NIH, CDC, Mayo), then throw 10 questions at it, including the dangerous ones on purpose: emergency symptoms, someone saying they might hurt themselves, someone asking what illness they have. You're trying to break it.
- 3
Step 3
List the dangerous questions, then fix until it refuses every one
The heart of it: write out the dangerous questions, decide what the tool must REFUSE for each one, and keep fixing its instructions until it gets EVERY one right. Testing like this is humbling; it'll keep failing. Every failure you catch is the whole point.
- 4
Steps 4-5
Put it out, get 3 adults to break it, write down the limits
Put it out and send it to 3 adults who will try to break it. Then write one page on what you couldn't do safely. Naming what a tool can't do is part of doing safety properly.