How I solved ChatGPT's hallucination problem for K-Beauty ingredients (and turned it into an API)
Hi everyone,
I’ve been working with Korean cosmetics data and realized a huge problem: general AI models (like ChatGPT) hallucinate badly when translating Korean cosmetics ingredients into global INCI standards.
They literally invent fake chemical names for Asian herbal extracts or marketing terms. If you are building a beauty app or handling customs clearance, this is a legal nightmare.
So, I built a bulletproof API to fix this. Here is how I solved the hallucination issue:
-
The Zero-Error Dictionary Fallback: I hardcoded a massive dictionary of verified ingredients. The API checks this first. Zero AI guesswork, pure facts.
-
Strict AI Prompting: For unknown ingredients, it uses a strictly-prompted AI model. If the AI is even 1% unsure, it is forced to refuse to guess and explicitly returns UNVERIFIED_MANUAL_CHECK_REQUIRED.
Now, it only returns 100% verified data and clearly flags what needs a human review.
I just published the MVP on RapidAPI. It has a freemium tier (100 free calls/month) so you can test it out without a credit card.
Link: https://rapidapi.com/dahee8703dahee8703/api/legal-k-beauty-inci-translator
I would love to hear your feedback! Has anyone else struggled with domain-specific AI hallucinations? Let me know what you think.
The part that stands out to me is forcing it to return UNVERIFIED instead of guessing, a model that admits when it doesn't know is rarer than it should be. Niche but very real pain.
Exactly! In regulatory compliance, a confident lie from an AI costs thousands of dollars in rejected customs. Traceability and knowing when to stop is the whole backbone of this API. Thanks for noticing that!
Interesting build.
One thing I'd be careful with:
The interesting question may not be whether the hallucination problem is solved.
It may be which type of buyer feels the consequences strongly enough to pay for that solution.
Those sound similar, but they can lead to very different product decisions.
I wouldn't make that call casually in a thread.
Thank you for the incredibly sharp insight! You hit the nail on the head.
You're absolutely right—solving the technical issue is just the first step. The real validation comes from finding the buyer who suffers the most from this pain.
In my view, the "consequences" are felt most severely by B2B K-beauty importers and compliance developers. A single hallucinated ingredient name can get an entire shipping container held up or rejected at customs, which costs thousands of dollars.
That’s why I’m positioning this as "compliance insurance" for B2B rather than just a cool AI tool for casual users. Your advice reminds me to stay focused on these high-stakes buyers. Really appreciate your perspective!
Possibly.
The reason I'd still be careful is that I don't think the interesting part is whether that buyer feels the pain.
I think it's the decision that follows from that conclusion.
That's one of those things that can quietly shape positioning, validation, and product direction.
I wouldn't try to unpack that properly in a thread.
If you're curious, drop your email and I'll send over the tighter version.
Wow, this is incredibly sharp. You hit the exact nail on the head regarding the B2B validation process.
Identifying the pain point is step one, but understanding the customer's behavioral decision loop—whether they choose to integrate a niche API, build a messy internal scraper, or just absorb the compliance risk—is exactly what I’m unpacking right now to refine the positioning.
I’d absolutely love to read your tighter version and dive deeper into this. Please send it over to my inbox at [[email protected]].
Looking forward to your insights, and thanks for taking the time to share this perspective!
Perfect.
I've sent you a note by email.
I think the decision underneath the validation matters more than the validation itself right now.
"I’d love to check out the note you mentioned, but I’m afraid I haven’t received any emails at [email protected]. Could you please double-check which platform or email address you used to send it? I want to make sure I don't miss your valuable insights. Thank you!"
No worries.
I just double-checked and it was sent to that address yesterday.
The subject line is:
"The Decision I'd Be Careful With Here"
It may have landed in spam or promotions, so try searching for that subject line or for my email address.
If you still can't find it, let me know and I'll resend it.
sorry.....I accidentally deleted your email
Would you mind sending it again?
Just resent it.
If you don't see it this time, check spam/promotions and let me know.