Skip to content
La Bonne Boite : Trouvez les entreprises qui recrutent régulièrement
TSQL Python Jupyter Notebook HTML CSS JavaScript Other
Branch: master
Clone or download

 _       _                            _           _ _
| | __ _| |__   ___  _ __  _ __   ___| |__   ___ (_) |_ ___
| |/ _` | '_ \ / _ \| '_ \| '_ \ / _ \ '_ \ / _ \| | __/ _ \
| | (_| | |_) | (_) | | | | | | |  __/ |_) | (_) | | ||  __/
|_|\__,_|_.__/ \___/|_| |_|_| |_|\___|_.__/ \___/|_|\__\___|

Build Status

Présentation du projet

Quel est le canal le plus utilisé par les chercheurs d'emploi pour rechercher un emploi ? ... Les offres d'emploi.

Quel est le canal le plus utilisé par les employeurs pour recruter ? ... Les candidatures spontanées.

Selon une enquête de l’INSEE, 7% des recrutements se font via des offres, contre 42% via des candidatures spontanées. Le « marché caché » (qui n’est pas matérialisé dans des offres) est donc la première source de recrutement en France !

La Bonne Boite (LBB) est un service lancé par Pôle emploi pour permettre aux chercheurs d’emploi de cibler plus efficacement leurs candidatures spontanées : l'utilisateur accède à la liste des entreprises à « haut potentiel d'embauche ». Le « potentiel d'embauche » est un indicateur exclusif inventé par Pôle emploi pour prédire le nombre de recrutements (CDI et CDD de plus de un mois) d’une entreprise donnée dans les 6 prochains mois.

En contactant des entreprises à « haut potentiel d'embauche », le chercheur d'emploi concentre ses efforts uniquement sur les entreprises qui sont le plus susceptibles de l'embaucher. La Bonne Boite lui permet ainsi de réduire drastiquement le nombre d'entreprises à contacter et d'être plus efficace dans sa recherche.

Le « potentiel d'embauche » est un indicateur basé sur une technique d'intelligence artificielle (apprentissage automatique ou "machine learning"), en l'occurence un algorithme de régression. Pour calculer un potentiel d’embauche, La Bonne Boite analyse des millions de recrutements de toutes les entreprises de France depuis plusieurs années.

La Bonne Boite a été déployée en France avec des premiers résultats encourageants, et est en cours de développement pour d'autres pays (Luxembourg).

La Bonne Boite c’est un site web mais aussi une API

La Bonne Boite, on en parle dans la presse

Project overview

A 2016 study by INSEE states that 7% of recruitments come from job offers, whereas 42% come from unsollicited applications. Thus the « hidden market » (not materialized in job offers) is the first source of recruitements in France!

La Bonne Boite (LBB) is a service launched by Pole Emploi (french national employment agency) to offer a new way for job seekers to look for a new job. Instead of searching for job offers, the job seeker can look directly for companies that have a high "hiring potential" and send them unsollicited applications. The "hiring potential" is an algorithm exclusivity created by Pole Emploi that estimates how many contracts a given company is likely to hire in the next 6 months.

By only contacting companies with a high "hiring potential", job seekers can focus their efforts only on companies that are likely to hire them. Instead of targeting every and any company that might potentially be interested by their profile, La Bonne Boite drastically reduces the number of companies a job seeker needs to have in mind when looking for a job.

The "hiring potential" is an indicator based on a machine learning model, in this case a regression. La Bonne Boite processes millions of recrutements of all french companies over years to compute this "hiring potential".

It has already been deployed in France with early results that are very promising. Early development is being made for new countries (Luxembourg).

La Bonne Boite is a web site and an API.

Press Coverage on La Bonne Boite

Table of contents


Install OS requirements:

On Debian-based OS:

$ sudo apt-get install -y language-pack-fr git python3 python3-dev python-virtualenv python-pip mysql-server libmysqlclient-dev libncurses5-dev build-essential python-numpy python-scipy python-mysqldb chromium-chromedriver xvfb graphviz htop libblas-dev liblapack-dev libatlas-base-dev gfortran

On Mac OS:

$ brew install openssl selenium-server-standalone
$ brew switch openssl 1.0.2s  # fixes error about libssl 1.0.0 missing
$ brew cask install chromedriver

On fedora

$ sudo dnf install python3-numpy glib docker-compose

You will also need to install docker and docker-compose. Follow the instructions related to your particular OS from the official Docker documentation.

Create a virtualenv for Python 3.6

For now, La Bonne Boite runs in production under Python 3.6.8. You are going to have to create a virtualenv that runs this specific version of Python.

$ wget
$ tar -xvzf Python-3.6.8.tgz
$ cd Python-3.6.8
$ ./configure --prefix=/usr/local --enable-loadable-sqlite-extensions
$ sudo make altinstall

Create an isolated Python environment, for example using virtualenvwrapper:

$ mkvirtualenv --python=`which python3.6` lbb
$ workon lbb

You might need to add labonneboite base directory to the Python path. This has to be run only once. One way to do it using virtualenvwrapper:

$ add2virtualenv $PWD

Install python requirements:

Our requirements are managed with pip-tools.

Note that you will need a local version of MySQL with mysql_config to install mysqlclient.

pip install --upgrade pip
pip install pip-tools
make compile-requirements

To update your virtualenv, you must then run:

python develop

How to upgrade a specific package

To upgrade a package DO NOT EDIT requirements.txt DIRECTLY! Instead, run:

pip-compile -o requirements.txt --upgrade-package mypackagename

This last command will upgrade mypackagename and its dependencies to the latest version.

Start required services (MySQL and Elasticsearch)

$ make services

Create databases and import data

$ make data

If needed, run make clear-data to clear any old/partial data you might already have.

Populate Elastic Search

This is required for autocomplete and search to work

$ make create-index

Known issues

You may have to run sudo usermod -a -G docker $USER, then reboot your computer to enable the current user to use docker, as the problem is described here

MacOS users, if you get a ld: library not found for -lintl error when running pip-sync, try this fix: ln -s /usr/local/Cellar/gettext/* /usr/local/lib/. For more information see this post.

Launch web app

make serve-web-app

The app is available on port 5000 on host machine. Open a web browser, load http://localhost:5000 and start browsing.

Run asynchronous tasks

Some parts of the code are run in a separate task queue which can be launched with:

make consume-tasks

Or in development:

make consume-tasks-dev

Asynchronous tasks are backed by Redis and Huey.

Run tests

We are using Nose:

$ make test-all

Access your local MySQL

To access your local MySQL in your MySQL GUI, for example using Sequel Pro:

  • new connection / select "SSH" tab
  • MySQL host:
  • Username: root
  • Password: leave empty
  • Database: labonneboite

You can also access staging and production DBs using a similar way, however with great power comes great responsiblity...


Access your local Elasticsearch

Docker forwards port 9200 from your host to your guest VM.

Simply open http://localhost:9200 in your web browser, or, better, install the chrome extension "Sense".

You can also use curl to explore your cluster.



# Cluster health check.
curl 'localhost:9200/_cat/health?v'

# List of nodes in the cluster.
curl 'localhost:9200/_cat/nodes?v'

# List of all indexes (indices).
curl 'localhost:9200/_cat/indices?v'

# Get information about one index.
curl 'http://localhost:9200/labonneboite/?pretty'

# Retrieve mapping definitions for an index or type.
curl 'http://localhost:9200/labonneboite/_mapping/?pretty'
curl 'http://localhost:9200/labonneboite/_mapping/office?pretty'

# Search explicitly for documents of a given type within the labonneboite index.
curl 'http://localhost:9200/labonneboite/office/_search?pretty'
curl 'http://localhost:9200/labonneboite/ogr/_search?pretty'
curl 'http://localhost:9200/labonneboite/location/_search?pretty'

DB content in the development environment

Note that we only have data in Metz region.

Any search on another region than Metz will give zero results.

Running scripts

For example create_index:

$ python labonneboite/scripts/

Running pylint

You can run pylint on the whole project:

$ make pylint-all

Or on a specific python file:

$ make pylint FILE=labonneboite/web/

We recommend you use a pylint git pre-commit hook:

$ pip install git-pylint-commit-hook
$ vim .git/hooks/pre-commit
# (...) previous content which was already present (e.g. nosetests)
# add the following line at the end of your pre-commit hook file

Debugging safely in a development, staging or production environment


# anywhere in the code"message")

# for an interactive debugger, use one of these,
# depending on which place of the code you are

# if you are inside the web app code
raise # then you can use the console on the error page web interface

# if you are inside a test code
from import set_trace; set_trace()

# if you are inside a script code (e.g. scripts/
# also works inside the web app code
from IPython import embed; embed()
# and/or
import ipdb; ipdb.set_trace()

Staging or production

Debugging safely in staging and in production is hard but not impossible! Here is how you can do.

Set a breaking point and target it

Add a breaking point in your code.

⚠️ This will of course block each incoming request, so make sure you know what you do.


import requests

def armstrong():
    if request.args('pdb'):
        from remote_pdb import RemotePdb
        RemotePdb('', 4444).set_trace()
    return redirect(url_for('root.home'))

Then target your breaking point making a request to your route. Example:

import requests

Nothing happens, it's normal! Time for step two.

Getting the IP address if you're using Docker

Connect to the server via ssh.

Then get the container IP:

$ docker inspect -f '{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' container_name_or_id
# to get the container name, do a `docker ps`

Connect to your breaking point with telnet :

$ telnet 4444
Connected to
Escape character is '^]'.

If you're not using Docker

Connect to the server via ssh.

Then enter your breaking point with telnet :

$ telnet 4444
Connected to
Escape character is '^]'.

Exiting pdb and telnet

To exit pdb, type 'c' in pdb and then ctrl+].

Then type quit to exit telnet.

(Pdb) c
# Then type 'ctrl + ]'
# And type 'quit'
telnet> quit
Connection closed.

About the API

For legacy reasons, the (LBB/LBA) API is not consumed by the LBB frontend, but it is consumed however by the LBA frontend, which is technically just an API consumer like any other.

API users must have an API key defined in the settings like this:

  'lba': 'dummykey',
  'emploi_store_dev': 'dummykey',

API users may have options defined in the settings like this:

  'lba': {
    'scopes': [Scope.COMPANY_WEBSITE, Scope.COMPANY_EMAIL],

scopes is an option used to let specific users access sensitive data.

Note for API proxies such as ESD (emploi store dev): the real user name needs to be forwarded in the GET param origin_user for each request. This will be taken into acccount to match a user to options in API_USERS.


The importer jobs are designed to recreate from scratch a complete dataset of offices.

Here is their normal workflow:

check_etab => extract_etab => check_dpae => extract_dpae => compute_scores => validate_scores => geocode => populate_flags

Use make run-importer-jobs to run all these jobs in local development environment.

Single-ROME vs Multi-ROME search

The company search on the frontend only allows searching for a single ROME (a.k.a. rome_code). However, the API allows for multi-ROME search, both when sorting by distance and by score.

How to contribute

For devs in the core team, this repo follows the Feature Branch Workflow.

We are also open to comments, questions and contributions from devs outside the core dev team! Here is a document about contributions to this project (in french)

Je Postule

Je Postule ("I apply") allows job seekers to apply directly and quickly to companies, which in turn can provide quick answers to applications. It can be integrated to many applications, like we did in La Bonne Boite.

In order to link your local instance of labonneboite with a local instance of jepostule, you will need to edit the file labonneboite/conf/ to override the settings of the section "Je postule" from the file labonneboite/conf/common/ In particular you will need to set the client ID and client secret provided by jepostule when you create a client platform as explained in the README section "Create a client platform".

How to disable JePostule on LBB frontend

When JePostule has serious issues (Mailjet issue and/or the whole service is unavailable) you want to hide the JePostule button on the LBB frontend to avoid frustrating your users.

You can do it pretty easily editing the JEPOSTULE_QUOTA setting:

JEPOSTULE_QUOTA = 0 # put 1 if you want to enable it.

One easy way to do that is to rebase and deploy this MR.

You can’t perform that action at this time.