Pet-sitter marketplace for France

The platform had to cover dozens of animal types — including birds and reptiles — each with different care requirements, risk profiles, equipment needs, restrictions, safety rules, and eligibility criteria. The data model had to absorb every plausible combination.

Sitter profiles carry a lot of operational data: availability, travel and hosting zones, animal-count limits, living conditions, experience, and restrictions. We had to build in validations, minimum requirements, and dependency checks.

A sitter could be free in general but not on the dates needed, refuse certain animals, or limit the format of care. The platform needed an availability engine that handles slots, exceptions, minimum and maximum durations, buffers between bookings, time zones, and partial occupancy — and keeps that logic synced with search and price requests, so the system never surfaces sitters who aren't actually available.

With this many variables, classic filters fall apart. We needed a ranking model that distinguishes must-have requirements from nice-to-have preferences.

Pricing across animals and care formats requires a lot of input. That meant a conditional questionnaire that captures enough data to produce an honest offer.

In real life, price depends on the animal, the number of pets, duration, season, location, and other factors. The product needed a model where sitters set ranges, rules, and surcharges, and the platform composes the final offer correctly.

Chat had to be tied to a specific request, with statuses, reply templates, attachments, reports, restricted contact-data exposure before a certain step, audit logs, and rate limiting.

If owners and sitters move conversations to outside messengers, the platform loses the transaction. Technical safeguards were needed to discourage that.

A clean booking → payment → commission → payout model had to be designed from day one.

A full moderation workflow was needed — queues, statuses, decision reasons — so support could work comfortably and the process was repeatable.

Managed data, reference dictionaries, roles, and product logic had to be split cleanly across the API layer.

Search that combines location, availability, dozens of filters, and ranking is computationally heavy. The service had to stay fast under that load.

The platform handles personal data. That brought technical requirements around data minimization, access control, audit logs, deletion and export flows, and retention policies.

We designed the entities and dictionaries — animals, care types, requirements, restrictions, zones, rules, statuses — along with the role model: owner, sitter, moderator, admin. We also built the support back office: moderation queue, complaint review, quick edits, blocking and hiding, action log. This lets the client evolve the product on their own, without rebuilding it every time.

We built a dedicated service responsible for baseline matching, availability, questionnaire rules, offer composition, dialogue statuses, anti-fraud signals, and audit. This keeps Directus from being overloaded and gives us a stable core that could be scaled and optimized independently.

The profile covers animal types (including exotics), care formats, conditions, limits (count, size, behavioral), pricing and add-ons, zones, availability, experience, and verifications. We saw the need for dependency validation and a profile-readiness indicator.

When an owner books care, payment is held by the system. Once the service is confirmed, funds are captured, the platform commission deducted automatically, and the payout released to the sitter's account. In cancellation cases, refunds follow the platform's rules.

We implemented an availability model with exceptions, slots, buffers, simultaneous-care limits, and conditional restrictions by animal type or care format. This lets search return only genuinely available sitters and cut down on broken arrangements.

Owners need to immediately understand which sitters are close and what the logistics look like. We implemented two synchronized search modes: a list view giving parameter-by-parameter comparison on cards, and a map view giving spatial context. Both modes share the same filters, radius, and availability logic.

We implemented ranking where critical requirements act as hard cutoffs, while everything else influences position in the list. We also built zero-result guidance (suggesting which constraint to relax) and comparison-friendly result cards, so owners could decide without opening dozens of profiles.

We built a structured, adaptive request flow — the question set adapting to the animal and care format (medications, behavior, equipment, routine, experience requirements). The output is a standardized data package ready for offer generation.

Messages, files, and statuses live inside a specific request with full history and event tracking. That preserves context, enables repeat bookings in one click, and leaves a clean path to future payments and bookings without rebuilding the foundation.

We implemented contact-data detection, suspicious-activity limits, risk signals (mass identical messages, spam patterns), controlled sanctions (warning, restriction, ban), and rules that only expose contact details at the right step. This protects both the monetization model and community quality.

We implemented statuses, queues, decision reasons, reply templates, moderator audit logs, and SLA rules for incident handling.

We integrated AI working on top of platform data to help owners reach the most relevant sitters faster. It analyzes a combination of signals from each sitter's profile — description, sitting history, animal types, exotic-pet experience, living conditions, limits, availability discipline, communication style, reputation markers — and uses them to surface relevant candidates in search.

to the first booking requests

to first contact between owner and sitter

processed on the platform before the investment stage

kept on-platform without moving to outside messengers

across all platform transactions