Inspiration
You've just landed somewhere new, or you're at the gym with your hands full, and you want to know where to eat, what to listen to, or what to do next. Search shows you what's popular, not what's yours. A friend who lives there would ask what you like and then answer in a sentence or two. Nion tries to be that friend: a voice that knows the place and knows you.
What it does
Nion is a voice-and-typing web app for people who are new to a place. A new visitor starts with a "Tell Nion what you love" card: pick artists, films or shows (Qloo suggests as you type), or tap music styles (pop, hip hop, jazz, electronic, indie, classical, metal) and kinds of film or show (comedy, sci-fi, drama, thriller, documentary, animation). Optionally, add a food and an age range. Say or type "I'm hungry" or "what would I like in Lisbon?" and Nion answers in a few spoken sentences while listing its picks on screen.
It picks places, music, films, TV, books, podcasts, brands, and destinations with Qloo. Your likes carry across all of these, so a music taste can drive place and film picks. Dislikes are more targeted: a disliked food only affects places, a disliked genre can affect any type, and a disliked named film or artist only affects the same type of thing.
Each pick has a "Qloo says" block showing whatever Qloo returned for it: the fit score, Qloo's description, which of your likes drove the pick and by how much, what a place is known for, and Qloo's tags. Places usually get all of it. Artists and destinations usually get just the fit score and tags. Under the picks, short lines say what Qloo matched, how many requests it took, and which model wrote the answer.
Where Qloo has opening hours for a place, the card shows whether it's open now and when it closes or opens, in the place's local time. Some places, such as the Brooklyn Bridge, have no hours, and their cards say nothing about it.
How to try it
A path to the strongest screens in under a minute:
- Open the link. On the card, type an artist, tap a style, and tap an age range, then press Start with my taste.
- Ask "What would suit me in New York?" You'll see the "Qloo says" block under each pick and the "tuned for ages…" heading.
- Press Forget my taste. This clears your likes and resets the card. Enter the same likes again, pick a different age range, and ask again. The picks change.
- Ask "Music like Radiohead", then "I love Blade Runner 2049. Where should I travel?" to see one taste crossing domains.
- Voice needs Chrome and a microphone. Typing works in any browser. Turn off Nion stops the mic.
The demo has limits to protect API credit, so nobody mistakes them for a bug:
- "Slow down" appears after more than 6 questions in a minute or 40 in an hour from one browser. Waiting about a minute clears the per-minute limit.
- "I'm resting" appears when the daily spending guard (about $2 a day, resets at midnight UTC), the total guard ($8, until the server restarts), or Qloo's monthly allowance is reached. If you see it, please try again later.
Why it only works because it knows what people love
- Same likes, same city, different age range, different places. Gallery image 2 shows it. The age range is a Qloo signal, so changing it changes the picks, and it isn't the chatbot guessing.
- One taste crossing domains. The taste comes from a different category than the question. A film taste answers a travel question, and a band answers a music question (gallery image 5). A band also gives places (gallery images 2 and 3).
How I built it
- You pick your loves on the first-step card, then speak (Chrome) or type, or tap a suggested example.
- The page sends your words, your saved taste, and your location to a small server that holds the API keys.
- Claude Haiku 5.5 is the agent. It can note a like or dislike, search the web once, or ask Qloo.
- Qloo returns picks. The server turns them into plain terms for Claude (a walking time, a fit in words), Claude writes the spoken answer, and the screen lists the picks Nion actually named.
How it uses Qloo
Qloo is the taste engine, not decoration. Nion uses three kinds of call: name lookup (/search), tag lookup (/v2/tags), and recommendations (/v2/insights). One question can trigger several, because each liked item and each tag is looked up separately. Qloo also powers the suggestions on the first-step card.
On the recommendations call, Nion uses:
signal.interests.entitiesandsignal.interests.tagsfor your likessignal.demographics.agefor the optional age rangefilter.tagsfor categories and cuisinesfilter.locationwithfilter.location.radiusfor the map pointfilter.exclude.entitiesandfilter.exclude.tagsfor your dislikesfeature.explainabilityfor Qloo's "why" breakdownbias.qualityfor rating-aware rankingoffsetto skip the results Nion has already named, so "more" gives new picks instead of repeats
Duplicates are removed separately, by de-duplication.
Nion respects both of Qloo's limits, by different means. It reads Qloo's monthly-allowance headers and rests before the allowance runs out. It stays under the 5-per-second limit by spacing calls 230 ms apart.
The age range is optional. It's one of three ranges, never an exact age. It's kept in your browser and sent with your question as a taste signal, and "Forget my taste" removes it. With a range set, the heading says "tuned for ages 36 to 55" (for example) and the picks change.
Nion asks Qloo for the "why" breakdown only when two or more likes are blended. With a single like it doesn't ask, because the answer would trivially be "all of it".
Who it's for
Nion is for anyone who needs good taste-matched answers without stopping to search and scroll:
- Travellers in a city they don't know, who want picks that fit them, not a list of what's popular.
- People whose hands or eyes are busy: carrying bags, at the gym, walking. Nion is built to work by voice, so hands-free use is possible wherever Chrome's speech recognition is available.
- People who don't want another account. There are no accounts and no database. Your taste is kept in your browser, and one taste works across eight categories.
Where this could go as a product: a hands-free companion that carries your taste from your commute playlist to your dinner to your next trip, without asking you to start over in each app.
Built as a complete product
- Typing works on a phone, and I checked the layout at phone width.
- A History page of your earlier questions.
- Four ways to take control: tap Stop, press Escape, tap the circle, or press Turn off Nion.
- "I'm resting" and "Slow down" lines instead of raw errors when a limit is reached.
- Hosted, and it will stay up through judging (Nov 16).
- MIT licensed.
Challenges
- Qloo's rate limit is the bottleneck. I assumed starting lookups in parallel would be faster. It wasn't, because Qloo allows 5 calls per second, which means at least 200 ms between calls. Nion spaces them 230 ms apart to stay safely under. Remembering lookups helped: the lookups before the recommendation call (names, tags, and cities) dropped from 1.4 s to 0.2 s. That is only that part, not the whole answer time. The lookups a demo is most likely to need (Radiohead, Blade Runner 2049, ramen, jazz, and three sample cities) are loaded in the background when the server starts.
- "Open now" had to be worked out. Qloo's hours filter takes only a day of the week, so I calculate open or closed from each place's hours, using the local time at the place. Later fixes: Nion no longer retracts hours after a "Thank you" (I ran that case three times live and it held). The model is also instructed not to mix up one place's hours with another's. That makes a mix-up much less likely, but it can't rule it out.
- A silent failure mode. My new tests caught a bug where a missing Qloo header could read as "0 calls left" and put Nion into "resting" mode. Fixed.
- Ranking quality. Places are ranked with ratings counting, and duplicates are removed. But the rating bias let a Thai place into a ramen list, so I turned it off whenever a cuisine filter is set. Qloo's own tags are loose, so a Thai place can still occasionally appear in ramen results.
- When the map service refuses. New York, Lisbon and Tokyo now work as sample cities even if the map lookup fails.
- Protecting credit. Per-visitor limits, daily and total spending guards, and a Qloo-allowance guard, with polite "I'm resting" lines.
What I learned
- Measure before optimizing. The obvious speedup (parallel calls) did nothing, and the boring one (remembering lookups) did the most.
- A voice product lives on a short path to a good answer. With a known taste, Nion goes straight to picks and says what it based them on. When it needs a location, there are one-tap buttons.
- Show the work. The "Qloo says" block and the who-did-what lines make it clear which part is Qloo's taste data and which part is the model's wording.
- Tests pay for themselves: 470 automated tests, real checks against Qloo and Claude, and a live-site test.
Honest limits
- Voice needs Chrome. I haven't tested voice on a phone, and Chrome on iPhone has no speech recognition, so typing is the way to use Nion there.
- Interrupting by voice is unreliable on speakers, so while Nion speaks the screen says to tap Stop or the circle to cut in, and there's a Turn off Nion button.
- Answers take a few seconds.
- Qloo's tags can be loose, and the "why it fits" sentence is partly Claude's.
- Places, music, film, and destinations were tested and tuned most. Books, podcasts, brands, and TV work through the same tool but weren't tuned.
What's next
- Support for browsers beyond Chrome
- More reliable voice interruption on speakers
- Tighter cuisine filtering
- Tuning books, podcasts, brands, and TV as carefully as places, music, film, and destinations
- Growing the hands-free, one-taste-everywhere idea into a travel companion