01
Proptech · Side project
2026-now
Side project, alongside client work · live product
Clavis.casa
Fewer wasted viewings, faster and safer long-term rentals.
Founder & Product Lead of my own side project. From discovery to live product in 8 weeks, entirely on my own.
Clavis.casa screens the profiles applying to view a rental property, validating identity and solvency before the viewing even happens.
Founder & Product Lead
Role
71 → 4 confirmed viewings
Pilot test
5 min, being reduced
Screening
Live product
Status
All product images are real screenshots, not mockups. The mobile ones are from the PWA, designed mobile-first partly for Google's indexing. Some parts of the process (the internal dashboard) stay private: Clavis.casa isn't a registered company yet.
Problem
The problem
Renting well on your own comes with real risks: the wasted viewing, renting below market value, or to someone who ultimately can't afford it. Handling the paperwork well still takes an agency or years of experience today, at the cost of high fees and lost time.
Understand
What I needed to understand
How to guarantee that whoever books the visit actually shows up, and that whoever shows up has their paperwork in order. I talked to real landlords to see how they handle this problem today.
Methods reported by the landlords I interviewed
- Personal details requested by message or form, before the visit.
- Direct replies to renters' own listing requests, offering the unit instead of a public ad.
- An ad with just the basics, full details only to whoever replies with their own information.
- Proof of solvency requested at the visit, like the employment contract I used to ask for myself.
Discover
What I discovered
I learned that the central problem, on both sides, is trust.
- The documents you receive make you responsible for their privacy (GDPR).
- No one validates the information they receive: more often than not, people lie just to get the visit.
- Candidates have a trust problem too: vague listings, with few photos and thin descriptions, are often the tool scammers use.
Change
What I changed
Rebuilding trust needed clarity on both sides:
- The landlord gets verified too, not just the candidate.
- The listing presents complete information.
- The processes make people aware of their privacy rights.
- Sensitive data is handled with total security.
These premises led to the three independent portals: landlord, candidate, and an impartial administrator, in a closed cycle.
Design decisions
The key choices behind every screen.
How it behaves
Real screen recordings of the product. The thread running through them: never leave someone alone in front of an empty field. Questions arrive one at a time, any answer can be corrected, and every photo brings its own prompts for describing it.
Validate
What I validated
Candidate side: a real pilot on the property listed on Idealista. Everyone who responded was invited to apply through the direct link to Clavis.casa, without being told they were part of a test.
Landlord side: a small demo campaign on a Facebook group for rentals in Turin, with a 45-minute video call for every interested landlord.
The underlying hypothesis, for both: if the information is clear, the right people come forward. Confirmed.
Result
The result
Candidate side: the screening works, 71 contacts down to 4 confirmed viewings. It currently takes 5 minutes.
Landlord side: of the demo requests, 1 in 3 turned into a completed video call, a high rate for such an early stage, with no marketing activity behind it. Fun fact: one person told me the process is better than Airbnb's.
I'm now in the go-to-market phase: SEO, a blog, and growing the number of landlords and agencies on the platform.
Candidate side · from contact to viewing
a qualification rate no manual check could match in the same time
The screening currently takes 5 minutes.
Landlord side · from request to demo
completed video calls, with no marketing activity behind it
One person told me the process is better than Airbnb's.
Brand
The brand
The typography follows the same idea: Fraunces for headings, the warmth of a classic serif; DM Sans for body text, the clarity of a technical sans-serif.
Terracotta
Sage
Ochre
Warm background
Design system
Design system
I defined reusable foundations and component variants to keep the interface consistent and scalable.
Below, the button system and tokens.
Motion
Motion belongs to the system as much as color, so it is documented here: whoever builds it needs to know what happens, not just how it looks. Three rules, each one born from a real problem.
Questionnaire
Single answer
- The tap is the answer: it confirms itself and the next question appears.
- Correct stays under every answer, so one can be changed without starting over.
Questionnaire
Multiple answers
- With several choices nothing can tell when you are done, so you say so with Confirm.
- That is the rule separating the two: where an answer has an end it confirms itself, where it has none it takes a gesture.
Photos and description
The photo slots
- They stay empty frames until something is uploaded.
- Once a photo is in, prompts arrive to help describe the room, and another room can be added.
What I learned
- A real test, even a small one, beats ten interviews: how people behave in front of an unfamiliar form is different from what they say they'd do.
- The screening works: going from 71 contacts to 4 confirmed viewings is a qualification rate no manual check could match in the same time.
- Being founder, designer, and, in practice, product lead of a team of AI agents requires a different kind of discipline: every design decision has to be clear enough to be translated into instructions for actually building it.
- Managing the project in Claude Code, after switching from Lovable, turned out to be more precise for this kind of product.
Need something like this?
Send me an email
