Skip to content
Burin

◐◐ or double-press and hold the page

Quote

01

All work

Case study / flashdebarras

Flashdébarras

A marketplace and operations tool for junk removal, cleaning and short-term storage — two brands on one admin, from first contact to invoicing, with an AI assistant that drives the admin panel in natural language.

Role
Design & build, end to end
Stack
Next.js · TypeScript · PostgreSQL · Prisma · Tailwind CSS · Claude API

02

Overview

The problem
Junk removal still runs on phone calls and spreadsheets. Requests arrive with no structure, volumes and access guessed at, and the handoff from estimate to job gets lost in scattered notes. On the storage side, no single place to know what’s kept, for how long, at what price.
The craft
A request flow that forces the right questions — property type, volume, floor and access, window — to price it right. Behind it, a dashboard that moves each job from quote to close, a storage module, and two brands — Débarras, Nettoyage — differentiated down to the quote and the invoice. On top, an AI assistant drives the admin in natural language — every action confirmed before it runs, customer data on a reversible soft delete.
The result
Day-to-day work shifts from a pile of threads and spreadsheets to a single job record per customer, legible from request to invoice. A real project, built end to end, still evolving.

03

On its real screens

  1. Professional clearance — the promise, up front

  2. The admin panel — the whole trade on one screen

  3. The AI assistant — driving the CRM in plain language

  4. Quotes, priced line by line

  5. The request, framed from first contact

  6. The public site, installable like an app

04

The admin, live

Two brands, one cockpit

One tool sells both clearance and cleaning: a quote switches brand with one control — pricing, deposit, terms and the PDF preview follow. And the job advances, state by state, on the same record.

Replayed continuously — the preview follows the brand.received → closed: the job advances

05

In detail

The context

Junk removal is fieldwork: clearing a flat, a cellar, a commercial space, sometimes under the pressure of an estate or a move. The invisible part — qualifying the request, estimating volume, scheduling the crew, billing — was being done by hand, thread by thread.

The request, framed from the start

The starting point was the request form. Instead of a free-text box, a flow that asks what matters for a quote: property type, estimated volume, floor and access conditions, preferred window. A complete request beats a fast one — it’s what removes the back-and-forth and the wrong estimates.

Operations

Each request becomes a job that advances through states — received, estimated, scheduled, done, closed. The dashboard gives the whole picture without hunting for information elsewhere.

Two brands, one tool

Clearance was only half the trade: the client also sells cleaning — end of lease, post-renovation, upkeep. Rather than dilute the brand, a second dedicated site lives on its own subdomain: “Flash Nettoyage”, a blue hero, call and WhatsApp one tap away, and cross-navigation between the two storefronts — each activity keeps its identity, each points to the other.

On the management side, no second tool: the same admin serves both brands, and everything in it is differentiated by service type. A quote or an invoice is created as “Flash Débarras” or “Flash Nettoyage” with a switch; pricing follows — flat fee or itemized lines —, the deposit is set as a percentage, the terms adapt to the brand automatically, and the PDF preview updates live as you type. Two commercial identities, one cockpit.

The admin, and the assistant that drives it

Behind the public site, an admin panel covers the whole trade: clients, jobs, quotes, invoices, finances, settings. It’s the cockpit — and it can also be driven in natural language. A built-in assistant understands “find the client Dupont”, “create a quote: cellar clearance 600 CHF, cleaning 300 CHF” or “rewrite the quote terms”, and prepares the requested action.

The rule that matters: the assistant proposes, the human confirms. Every write — a quote created, an address changed, a job closed — is shown before it runs, because a model should never write into real customer data on its own. The comfort of natural language, the safety of systematic confirmation.

The admin installs like an app and fits in a pocket: on site, the crew reads the job and moves it forward from a phone.

Storage

The storage side ties a stored lot to a customer, a duration and a billing line, answering the simple but recurring question: what’s with us, since when, and until when.

Stance

A real project, built end to end and still evolving. No headline numbers here: what counts is that fieldwork now runs on a single, legible job record instead of a stack of tools.