donnee.dev

Websites, apps and AI assistants that do the work.

Donnée is a development and AI studio based in Montpellier, France. We design, we ship, we maintain. In French, English, Russian and Italian.

5 products in productionsites, platforms, real-time systems
From 2 days to 6 monthsfixed price, day rate or maintenance
Reply within 24 hoursone contact, from the first call to support
Studio availability50 % of capacity booked over the next 4 weeks
11 free hours per week Next slot for a new project: Fully booked for now, waiting list open

What we build

Six ways to help. Each service has its own shape in the field of dots.

A site that reassures, an application that holds up, an assistant that answers correctly. We always start with the question that has a number in it, and we end with something in production that you can evolve without us.

  • Websites and web applications

    From a firm's website to a full SaaS. Fast on phones, readable by Google, GDPR-ready.

    • Showcase and corporate sites, in Next.js or WordPress depending on the need
    • Online stores: WooCommerce, Stripe, payments and shipping
    • Business applications and SaaS: TypeScript, Postgres, customer areas
    • Portals, extranets, dashboards
    • Multilingual (i18n, hreflang), technical SEO, accessibility
    • Browser extensions
    ToolsNext.js 15, TypeScript, React, Tailwind, PHP, WordPress, Postgres, Drizzle
  • AI assistants

    They read your documents, use your tools and know how to say "I don't know".

    • Personal assistant: email, calendar, notes, on your own data
    • Customer assistant on your site: answers, qualification, appointment booking
    • Internal assistant: search across your documents, procedures, knowledge base (RAG)
    • Document generation: quotes, reports, PDFs
    • Agents connected to your tools: MCP, API, CRM, invoicing
    • Half-day AI workshops for your team
    ToolsClaude API, Anthropic SDK, RAG, MCP, OpenAI, Supabase
    See the AI assistants page
  • Machine learning and data

    One question with a number in it, one simple model, one answer measured before it is believed.

    • Forecasting and scoring: price, demand, risk, lead time
    • Classification of documents, emails, tickets
    • Anomaly and drift detection
    • Statistical validation: backtests, confidence intervals, rejection thresholds
    • Data pipelines, PostGIS, maps and dashboards
    ToolsTypeScript, Postgres, PostGIS, MapLibre, model APIs
  • Automation and integrations

    The tasks you redo every week, done once and for all.

    • Connecting tools through APIs and webhooks
    • Back-office flows: quotes, invoices, reminders, notifications
    • Payments (Stripe) and transactional email (Resend)
    • Automatic processing of incoming documents
    • Job queues and scheduled processing
    ToolsStripe, Resend, pg-boss, webhooks, REST APIs, Zod
  • Real-time systems and infrastructure

    What has to run at night runs at night, monitored and backed up.

    • Real-time connectors: WebSocket, market feeds, sensors
    • VPS servers, Docker, deployment on Vercel and Supabase
    • OVH hosting, DNS, certificates, email
    • Monitoring, alerts, backups, incident recovery
    ToolsVPS, Docker, Vercel, Supabase, OVH, WebSocket
  • Support, rescue and improvement

    Hacked site, vanished provider, code nobody dares to touch: we take over.

    • Maintenance and fixes: monthly plan or on demand
    • Taking over a hacked, broken or abandoned site
    • Performance, security, updates, GDPR compliance
    • Migration: WordPress to Next.js, legacy PHP to modern
    • Technical audit and AI readiness check: one day, written report
    ToolsWordPress, WooCommerce, PHP, MariaDB, Next.js, OVH

Live

A studio that talks about data should show some. These numbers come from open sources, right now, with no key and no cookie.

Each tile queries a public source when you open the page. If a source does not answer, the tile says so; it never makes anything up. That is exactly what we expect from your dashboards.

Electricity in France, now

RTE éCO2mix, real-time data

Querying

CO₂ in the atmosphere

NOAA via global-warming.org, daily mean

Querying

Earthquakes, last 24 hours

USGS, magnitude 2.5 and above

Querying

International Space Station

Where the ISS at, current position

Querying

Space weather

NOAA SWPC, planetary Kp index

Querying

Euro

European Central Bank via Frankfurter

Querying

Montpellier

Open-Meteo, conditions and air quality

Querying

World population

Estimate computed from UN 2024 projections, not a measurement

Querying

This page, measured in your browser

Web Vitals, Navigation Timing

Querying

AI assistants

A useful assistant knows your data, can use your tools and knows when to hand over.

For you

An assistant that reads your email, keeps your calendar and finds your notes, on your data, not a third party's.

For your business

An assistant that answers your customers, prepares your documents and searches your files, with rules, logs and a person as backup.

See the AI assistants page, with a demo

Machine learning and data

A model is worth building when it answers a question with a number in it: which price, which risk, which lead time, which customer.

We do not sell models for their own sake. We start from the data you already have, measure what a simple model adds, and only ship what survives validation.

  • Forecasting prices, demand, lead times
  • Risk and priority scoring
  • Classification and automatic sorting
  • Anomaly detection
  • Statistical validation and backtests

Already in production

Property price estimation from notarised sales

Confidence index, dispersion, outliers and a map of sales, built on French DVF data. Postgres, PostGIS, MapLibre.

Statistical certification of trading strategies

Backtests, rejection thresholds and continuous execution on a terminal that runs 24/7. TypeScript, exchange connectors.

Documents generated by a language model, checked against a schema

Structured PDF generation through the Anthropic API, with field validation before rendering. PHP, MariaDB.

Work

Five projects in production. Clients stay anonymous until they agree to be named.

  • Quantitative trading terminal

    Real-time connectors to two exchanges, statistical certification protocol, continuous execution on a VPS.

    Studio productTypeScript, WebSocket, VPS, DockerRunning 24/7 since [to confirm]
  • Multilingual platform for a condominium council

    Document and resident management, AI-generated PDFs, interface in several languages.

    Client, full deliveryPHP, MariaDB, Anthropic APIVersion 4.2 in production
  • Multilingual site for a consulting firm, with an AI intake assistant

    Next.js site in three languages, contact assistant powered by Claude, CNIL-compliant.

    Client, full deliveryNext.js, TypeScript, Claude API, Vercel[to confirm]
  • WooCommerce store recovered after a hack

    Cleanup, data restoration, hardening of the shared hosting, back online.

    Client, emergency workWordPress, WooCommerce, PHP, OVHBack online in [to confirm]
  • Prix du Quartier, property estimation

    Public estimate from notarised sales, seller cabinet with CRM, buyer portfolio and browser extension.

    Studio productNext.js 15, PostGIS, Drizzle, Stripe, SupabaseLaunching

How we work

Five steps, always the same. You know where the project stands at any time, and what is left to pay.

  1. Scoping

    One call, a written brief with the problem in your words, a quote that says what is in and what is out.

  2. Build

    A demo every week, on a real address. A decision log: what we chose, what we rejected, why.

  3. Delivery

    The code, the documentation, a handover session and the honest list of what remains open.

  4. Review

    Within 48 hours: measured results, what took longer than planned, what you can reuse elsewhere.

  5. Support

    Maintenance, improvements, on-call if the system runs continuously.

Formats and rates

Three ways to work together. The day rate is the same as on Malt, no surprise between the two.

Fixed price

Written scope, fixed price, delivery date. For a site, an application or a first assistant.

On quote, estimate below

Day rate

One or more days a week, on your backlog, with your team.

550 € per day

Maintenance

Monthly plan: updates, backups, monitoring, fixes, small improvements.

from €150 a month

Quick estimate

A range in thirty seconds, computed with our real rates. The quote comes after a scoping call.

The calculation starts from 92 € per hour of production. A billed day at 550 € covers 6 hours of actual work; the rest is coordination we do not bill.

Project type
Size
Options
Conditions
Where the hours come from
hours of production
billed days
price range
indicative timeline

Indicative range, VAT not applicable, art. 293 B of the French tax code. The firm quote follows a free 30-minute scoping call.

The studio

Donnée, in Montpellier. Small by choice, direct by principle.

You talk to the people who build, not to a project manager relaying messages. The studio's founder ran a web agency with developers and designers before coming back to code, and studied economics and law before that. It helps to scope a project, price it and read a contract.

We ship code we can explain line by line, and we would rather say "this will take three more weeks" than let you find out at delivery.

LocationMontpellier, remote or on site in Occitanie
TermsInvoicing in France, written contract, GDPR
LanguagesFrench, English, Russian, Italian
SinceIndependent since 2021

Let's talk about your project

Describe the problem in three lines. We reply within 24 hours with questions or a proposal.

contact@donnee.dev
or Malt profile

By sending this form you agree that this information is used to reply to you. No other use.