Lumen - medicine, made legible

Hey Product Hunt 👋 I built Lumen because of a moment I think most people have had. It's late, something hurts, and you don't know if it matters. You don't want to wake anyone over nothing. You don't want to sit in a waiting room until dawn over nothing. So you search it—and somewhere around the third result is the word you were afraid of. Now you're not sleeping either way. Then it happens again in two other places. You leave the pharmacy with three boxes and a leaflet folded into eighths; one warns about drowsiness, and another in smaller print does the same, and nobody mentioned what happens when you take both. And your lab results land in the portal on a Friday afternoon with four values flagged red, your appointment is Tuesday, and none of them are words you've ever used. Lumen is three tools for those three moments. 📷 Photograph your medicine boxes. What each one is for, and which combinations are worth asking a pharmacist about. 📄 Paste your lab results. Every value in plain language—including the ones that were completely normal. It leads with the count: "12 within the ranges on your report, 2 worth discussing, 1 outside." Most tools show you three red flags, and you leave convinced your whole body is failing. 🩺 Describe a symptom. How urgently to seek care, what would change that answer, and where the nearest real emergency department actually is. You pick how much you already know, and it explains at your level—from "ferritin is the pantry, not the plate" up to the pharmacology. The part I actually care about: Lumen shows it's working. Every AI health tool asks you to trust it. This one doesn't. When Lumen flags a drug interaction, it quotes the sentence from the official FDA label, verbatim, and names the drug. If it can't find a supporting sentence in the retrieved label, it doesn't make the claim. An automated test fetches the live FDA API and checks that every quote is genuinely present. No AI judging AI—just a string match. A fabricated interaction is the failure that could actually hurt someone, so I made it mechanically detectable instead of a matter of trust. Same principle throughout: lab reference ranges come only from your document, never the model's memory, because ranges vary by lab and assay. Clinic locations come from real map data, because a hallucinated hospital is someone driving somewhere that doesn't exist. And what it refuses to do: it won't tell you what you have. It can't examine you, and guessing is how people talk themselves out of care they need. It never tells you to start, stop, or change a medication. It stores nothing—no account, no history. None of this replaces your doctor. It's about not spending the hours before you see them frightened by things nobody took the time to explain. I'd genuinely like to know where it's wrong—a medication it can't resolve, a lab format it misreads, a triage answer that felt off. Those reports are worth more to me than upvotes. And a question for this crowd, because I went back and forth on it: was refusing to diagnose the right call, or does it make the product less useful than it should be? I think the constraint made it better. Curious whether you agree.