Hippo Homes

Your AI apartment hunter for Paris. Unifying live listings, learning your preferences, and turning the rental search into personalized, actionable matches.

Comment

Inspiration

Finding an apartment in Paris can be really tough. The good ones get taken quickly and old listings stay up forever. Every website has information, which makes it hard to find what you want. Hippo Homes was started because we asked a question: what if you had a personal agent who helped you find an apartment?

What it does

Hippo Homes puts all the Paris rental listings in one place. Checks them all the time. It learns what you like and dislike. Gives each home a score based on how well it fits you. You can compare apartments based on price, location, size and other things. You can also keep track of your applications, and enable the auto-application mode, that will submit your application to apartments that are a great match to your preferences.

It also helps owners by letting them list their apartments directly and by giving them tools to manage leases and payments.

How we built it

he product is a Next.js and React application backed by Supabase and Firebase App Hosting. Source-specific TypeScript connectors collect and normalize rental listings from Paris agencies and portals. A lifecycle engine audits availability, detects stale listings, deduplicates inventory, and preserves exact source attribution.

The recommendation layer combines deterministic constraints (e.g. size, price) with learned preferences. Image analysis extracts visual signals such as natural light, condition, room character, presence of mold, and kitchen quality. Those signals join budget, surface, area (arrondissement), floor, DPE (energy consumption), and amenities in an explainable scoring pipeline. Google Maps powers the geographic experience, and automated Vitest coverage protects ranking, listing lifecycle, mapping, applications, privacy, and reliability behavior.

Codex and more specifically GPT 5.5 and 5.6 were core development partners during this Build Week: they helped turn product ideas into implementation plans, accelerate TypeScript work, debug deployment and integration issues, extend test coverage, and harden the experience across the full stack.

Challenges we ran into

The hardest challenge was data trust. Listing sites use inconsistent markup, incomplete addresses, changing galleries, and pages that may remain accessible after a home is gone. We built source-aware extraction, evidence-based geolocation, conservative availability rules, and graceful fallbacks rather than inventing missing facts.

A second challenge was keeping AI useful and accountable. We separated hard constraints from behavioral signals, exposed reasons behind every score, tracked confidence, and made inferred preferences editable. Also, sometimes, an element that was functional in one version became dysfunctional in the following one, so it was very important to check what GPT was doing.

Accomplishments that we're proud of

  • A working Paris rental dashboard with live, source-linked inventory
  • Personalized, explainable matching across hundreds of user interactions
  • An AI “search DNA” (preference prediction) that adapts without overriding explicit preferences
  • Reusable renter application information, in spite of the application tracking that still needs to be built
  • A polished public experience with privacy, cookie, and ranking transparency
  • A production-oriented architecture with tests, quotas, locks, and degraded modes

What we learned

AI delivers the most value here when it is given a prompt that reduces uncertaint, because when there is room for mystery, the outcome can be very surprising, and rarely in a positive way. So it is compulsory to implement Human-In-the-loop principle, and never just let AI figure out everything. We also learned that narrowly focusing on Paris improves data quality and makes the experience far more coherent than chasing geographic scale too early. A lot was learned about compliance related to data in Europe as well, as the laws are very strict.

What's next

Next we plan to deepen source partnerships to allow including websites that do not allow data collection from their websites, implement the auto submission to all rental agencies websites, build the application tracking system, add verified application receipts, improve image understanding, expand the 3D Paris exploration experience, and pilot owner workflows with a small group of independent landlords and agencies. The long-term goal is an end-to-end rental journey, from discovery to a signed lease and transparent rent management, without selling renter data or charging search fees.

Built With

Share this project:

Updates