My daughter has epilepsy. She also has multiple other disabilities, which means that managing her safety isn't just about medication — it involves understanding how the environment she moves through might affect her. What she eats. What cleaning products are used in her home. What toiletries are in the bathroom.
Most people don't know that certain chemical compounds and essential oils can trigger seizures. Not just when ingested, but in some cases through smell alone. Eucalyptus is one of the most well-known examples. There are many others. The standard approach for families like mine is to crouch in a supermarket aisle, squinting at a tiny label, cross-referencing against a list you're trying to remember, hoping you haven't missed anything.
It's exhausting. And it's a problem that, on its face, AI should be able to help with.
Why most "helpful" AI tools aren't.
There's a particular failure mode common in AI tools built for people with complex health needs: they're designed by people who don't have those needs, for a user they've modelled rather than encountered.
The result is usually something that technically works — it can answer questions, retrieve information, present data — but that creates friction in the exact moments when friction is most damaging. When someone is in a supermarket with a limited window of time and cognitive load. When someone is exhausted, because managing disability is exhausting. When someone needs a clear answer, fast, without needing to interpret a probabilistic output or caveat-laden response.
Lived experience changes the product requirements. Not superficially — not just the language you use or the accessibility features you add — but fundamentally, at the level of what the product is actually trying to do.
What we built.
EPCheck is a free community safety tool. You open it in your browser — no download, no account required. You type a product name or upload a photo of the ingredient label. EPCheck checks the ingredients against a database of known and suspected seizure triggers and returns a result within seconds.
Each flagged ingredient comes with context: the risk level, how exposure occurs (topical, inhalation, or ingestion), what the evidence shows, which common products contain it, safer alternatives, and how quickly a reaction might occur. It's not just a red flag — it's enough information to make a decision.
The design decisions that came from living it.
Speed over completeness. A tool that gives you a preliminary result in two seconds, with the option to dig deeper, is more useful in a supermarket than one that gives you a comprehensive report in thirty. We optimised for the use case, not for demonstrating capability.
No account required. Requiring registration to access a safety tool creates a barrier at exactly the moment when someone needs help most. EPCheck works immediately, for anyone, with no registration. That was a non-negotiable.
Context, not just a flag. Telling someone "this ingredient is risky" without explaining why, how risky, or in what form of exposure, creates anxiety without agency. We give people enough information to make their own decision.
Community knowledge is valid knowledge. Peer-reviewed research is the foundation. But the epilepsy community has accumulated enormous amounts of practical knowledge about what triggers seizures that hasn't made it into published research. EPCheck treats community feedback as a legitimate input, not a lesser category of evidence.
What this means for how you build AI tools.
EPCheck is one example of a broader principle: the most useful AI tools for communities with complex, high-stakes needs are the ones built closest to the people who will use them. This doesn't mean every AI tool needs to be built by someone with the specific condition or need it serves. It means the people who will use the tool need to be involved in defining what it should do.
It means designing for the actual use context — which, for many accessibility and health applications, is high cognitive load, high emotional stakes, and limited time — not for the polished demo scenario. It means treating community knowledge as legitimate input. And it means being willing to constrain what the tool does. The temptation in AI product development is to expand capability. The wisdom is in knowing what to leave out.