Case studies

Real clients. Different industries. Measurable outcomes.

No generic testimonials. Delivered projects, with real context, the stack we used, and a measurable gain for the founder.

−24
days/yr recovered
Delivered
Sector : Carpentry / Wood flooring

Pierre et Parquet

Mathis Cottier, owner

Context

Pierre et Parquet imports its product catalog into Odoo manually every month. Two full days absorbed by entering supplier references, prices, stock and technical sheets.

Problem

Manual Odoo entry of 2 days/month for 800 to 1,200 references. Price errors leaking into customer orders. No margin to expand the catalog without hiring.

Solution

AI agent for catalog import: upload supplier file (PDF or Excel) → AI extracts references, normalizes prices, maps Odoo categories, then pushes products via the Odoo API after a one-click human validation.

Stack

  • Next.js
  • Gemini 2.0
  • Supabase
  • Vercel AI SDK
  • Odoo API

Outcome

2 days/month → 0 minute. 24 days per year recovered from manual entry.

« We had a wall. Every month two days locked updating Odoo, and no way to grow the catalog without breaking everything. Codea delivered the agent in three weeks, and I haven't opened the import screen by hand since. »

Mathis Cottier, owner, Pierre et Parquet
≈10 h
per week recovered
Delivered
Sector : Food / Quick-service

Dream Donuts

Thomas, founder

Context

Dream Donuts drives sales through several channels (POS, Uber Eats, Deliveroo, click & collect). Every Monday, Thomas spent a day compiling numbers into a spreadsheet for the team meeting.

Problem

One Monday/week blocked on reporting. Data scattered across 4 sources, manual copies, Excel formulas breaking each time a channel was added.

Solution

Self-hosted n8n workflow that queries each source API, aggregates sales by channel and product, asks Gemini for a synthetic commentary on variations, then sends a structured Telegram message to the founders every Monday at 8 AM.

Stack

  • n8n self-hosted
  • Gemini 2.0
  • Telegram Bot API
  • Supabase logs

Outcome

8 to 10 hours/week recovered. Reliable reporting, comparable week-over-week, no one opens Excel anymore.

« Monday morning, the Telegram message arrives. I know where we stand in two minutes. That's exactly what I wanted from a custom tool: it works for me, not the other way around. »

Thomas, founder, Dream Donuts

Have a workflow that costs you a day every week?

A free 30-minute audit to spot the first candidate worth automating.

Book my free audit