The engine behind JaxHomes
Taste Recognition, Analysis and Visual Intelligence System
TRAViS is a predictive preference engine. It watches how you react to real homes, builds a picture of your taste, predicts what you'll love next, then checks its own work and fixes what it got wrong. Around a hundred likes and dislikes is usually enough to get useful results; after that, confidence, granularity, and specificity keep improving the more you use it, as long as your reactions stay honest and consistent.
JaxHomes started inside a brokerage, watching the same thing happen over and over: a buyer sends their agent twelve listings, the agent sends back twelve more, and nobody can say out loud what actually separates a “yes” from a “no.” The search criteria were always right. The homes were always almost right.
What worked in person was simple: walk a few houses together and an agent starts to know. She stops sending the dark split-levels. She flags the one with the courtyard before you ask. That instinct is pattern recognition, built from watching reactions, and it has never existed online.
TRAViS is that instinct, made into software. Not a chatbot asking what you want, and not a filter bank. A quiet system that learns the same way a great agent does: by paying close attention to what makes you stop scrolling.
A recommendation list ranks what's popular. A preference engine models one person. A predictive preference engine goes one step further: it commits to a guess about what you'll do next, then measures whether it was right and pays for being wrong. That accountability loop is what turns taps into accuracy.
Every tap, swipe, lingering scroll and skipped photo is a signal. You never fill out a preference form; you just react to real homes.
TRAViS scores the next home before you see it: how likely you are to save it, how confident it is, and which traits drove the call.
Then it grades itself. Predicted a match and you passed? That's a miss, and misses are the most valuable data in the system.
Confident misses trigger a larger correction than close calls, so one surprising home can reshape the model while a coin-flip barely moves it.
Accuracy is tracked continuously. Signals that stop predicting well lose influence on their own, so the loop self-heals without anyone retuning it.
The best homes rise to the top of your deck, the sure misses sink, and uncertainty gets sampled on purpose, so every session teaches more than the last.
Rate a few homes. Watch the prediction sharpen, and correct itself when it's wrong.
Photo 1 · Number of stories
No signal yet: the first guess is deliberately neutral.
four stacked townhome levels.
Nothing yet. Rate a picture and the traits it shares will start showing up here.
Nothing here was asked in a form. Every bar came from taps alone.
-
All ratings
-
Last 5 ratings
-
Your consistency on repeats
Green ✓ means TRAViS predicted your choice. Red ✕ means your choice surprised it.
↻The small repeat marker appears only when TRAViS re-tests a strong prediction using an earlier picture. It is separate from whether the prediction was right.
These are the same practice pictures the guided setup uses. After enough ratings, an earlier picture may return when the model has a strong prediction to verify, not on a fixed cadence. This is a simplified illustration, not the live engine. The real system reads the photos themselves, weighs recency and hesitation, separates taste from budget, and blends thousands of signals per member.
Beds, baths and price narrow a list. They never explain why two homes at the same price feel completely different. TRAViS learns the feel: light, proportion, finish level, outdoor life, how a kitchen reads in a photo.
Ask someone what they want and you get a wish list. Show them forty rooms and you get the truth. TRAViS is built around reaction, not self-report.
Three weeks into a search, most buyers want something different than they did on day one. Recent signals count more than old ones, so the model follows you instead of anchoring to your first session.
A system that only confirms what it already believes stops learning. TRAViS deliberately mixes in homes it isn't sure about, because uncertainty is where the next insight lives.
The loop above is honest, but it's the outline, not the recipe. How TRAViS reads a photograph, which signals it trusts when they disagree, how it separates taste from budget and timing, how quickly influence decays, and how it stays useful for a brand-new member with almost no history. That part stays in the hive.
Your ratings are yours. TRAViS uses them to serve you better homes and to help your agent show up prepared, not to sell your taste to anyone else.
Rate enough real listing photos and TRAViS starts putting the right homes in front of you. Keep going, and confidence, granularity, and specificity keep improving as long as your reactions stay honest and predictable.
Start training TRAViS