One platform, across sectors

Healthcare, employment, social services and nonprofits, governments: the sectors where we deploy the Bayes Platform, and what it already does there.

Our approach

  1. The shared problemProfessionals supporting ever more people, under rules that keep changing.
  2. Our answerThe Bayes Platform, deployed with the institutions and nonprofits of each sector.
  3. For a governmentDigital public infrastructure: one shared foundation on which every ministry builds its agents.
Use cases

What gets built on the Bayes Platform

The agents our partners build or study on the Bayes Platform, by type.

44 use cases

  • Healthcare

    Augmented patient portal

    In the hospital's patient portal, the agent finds the right procedure among many and brings it forward.

  • Healthcare

    Therapeutic patient education

    Patients ask about their treatment between two appointments, from a corpus their care team has validated.

  • Healthcare

    Patient pre-consultation

    Gathering and structuring what the patient says before the first consultation, and spotting emergencies.

  • Healthcare

    Post-discharge patient follow-up

    Collecting the patient’s symptoms at home and passing the signals that matter back to the ward.

  • Healthcare

    Chemotherapy pathway assistant

    From the patient interview to the discharge letter, with anomalies flagged to the prescriber.

  • Healthcare

    Conversational welcome booklet

    Patients query their ward's welcome booklet; care questions are sent back to the team.

  • Healthcare

    Pre-operative information for families

    Reliable answers for parents, from the anaesthesia consultation to the day of surgery.

  • Healthcare

    Home-care companion

    Answering, day and night, the family caregivers who look after someone at home, and spotting emergencies.

  • Healthcare

    Parent psychoeducation after a diagnosis

    Supporting parents through the weeks after their child’s diagnosis.

  • Healthcare

    Assisted replies to patient messages

    The team answers patient messages faster, drawing on its past replies.

  • Healthcare

    Personalised health prevention

    Sourced prevention advice, fitted to the person’s health and where they live, with the right places to go.

  • Healthcare

    Mental-health orientation

    Informing people, young people first, and pointing them to the right medico-social services.

  • Healthcare

    Health education for teenagers

    Answering, without judgement, the questions teenagers do not dare to ask.

  • Healthcare

    Clinical protocol navigator

    Finding the right protocol, guiding its steps and showing the source.

  • Healthcare

    Patient-record summary

    A scattered patient file becomes a sourced, dated summary for the emergency doctor.

  • Healthcare

    Prescription assistant

    From the patient’s profile to a reasoned prescription, on the department’s own references.

  • Healthcare

    Lab-results conclusions

    Structuring each sample's data and drafting a conclusion that follows the lab's quality rules.

  • Healthcare

    Clinical-trial eligibility pre-screening

    Reading a patient file, assessing trial eligibility, flagging missing data and justifying the answer.

  • Healthcare

    Care-offer navigator

    The professional describes the need; the agent asks, searches the directories and answers with a map.

  • Healthcare

    Clinical-guidelines knowledge base

    Querying official guidelines and getting an answer that cites the paragraph.

  • Healthcare

    Health data in natural language

    Clinicians query a patient's data, connected devices included, without writing a query.

  • Healthcare

    Adverse-event reporting

    Guiding staff through the report, collecting complete information and giving a first response.

  • Healthcare

    Care-quality indicator analysis

    Rebuilding the care pathway from coded data and reading the indicators against national benchmarks.

  • Healthcare

    Systematic-review screening

    Screening articles against a review's inclusion criteria, with the reason for each decision.

  • Healthcare

    Coding clinical reports for research

    Extracting from a clinical report the scores a study used to code by hand.

  • Healthcare

    Regulatory assistant for clinical research

    Turning a regulatory requirement into a protocol decision, with the texts behind it.

  • Healthcare

    Symptom triage and outbreak monitoring

    Steering someone with symptoms to the right level of care and preparing what the clinician must report.

  • Employment

    AI coach for jobseekers

    In their own personal space, a coach that points to the right service, estimates entitlements and finds openings.

  • Employment

    Job and support search for caseworkers

    From the assistant they already use, caseworkers find openings, hiring companies and local services.

  • Employment

    Youth-employment adviser assistant

    Finding the local support, training and partners for a young person, in the adviser's own documentation.

  • Employment

    Personalised action plan for young people

    A young person's answers become a checkable action plan, drawn from a directory the team keeps up to date.

  • Employment

    Career discovery for a sector

    Discovering the careers of a sector, care and social work for instance, and the training that leads there.

  • Employment

    In-demand jobs explorer

    Browsing the in-demand jobs of a region and the training that leads to them.

  • Employment

    Assisted directory update

    An agent prepares the yearly update of an official list; a person approves it.

  • Employment

    WhatsApp assistant for entrepreneurs

    Answering supported entrepreneurs on WhatsApp, from the training content.

  • Social & Nonprofits

    Social worker copilot

    Notes that become action plans, case files that keep themselves up to date.

  • Social & Nonprofits

    Knowledge agent for advisers

    Making searchable the answers social workers keep, situation by situation.

  • Social & Nonprofits

    Benefits navigator for families

    Telling a family which support it is entitled to, and guiding it through the steps.

  • Social & Nonprofits

    School-family mediation

    Easing exchanges between schools and families who don't speak French.

  • Governments

    Conversational public-service counter

    A procedure handled end to end in a messaging app: the need, identity, a pre-filled form, payment.

  • Governments

    Case officer copilot

    The request lands in the officer's tool, where they see what the AI did, on which sources, and approve it.

  • Governments

    Archive digitisation

    Paper archives turned into a structured, searchable base.

  • Governments

    Business tools for a public assistant

    Building and maintaining the tools a public assistant calls, one per use case.

  • Governments

    Evaluating a deployed generative AI

    Measuring the quality of an assistant already in place, with the platform's evaluation tools.

Healthcare

Use cases built by caregivers

Impulse Healthcare, the programme

Impulse.HealthcareWith support from Google.org

Launched in January 2026 in partnership with AP‑HP, Paris, Impulse.Healthcare is how the Bayes Platform entered public hospitals.

Hospital teams turn their ideas into healthcare use cases they build themselves on the platform, tested in real conditions and shared as open-source digital commons.

Supported by a $5M grant from Google.org, the program is now expanding across Europe, with cohorts launching in Italy, Germany, Poland, and Spain by end of 2026.

23

Healthcare use cases built by caregivers

15+

Hospitals that built one of them

200+

Caregivers trained

$5M

Google.org grant

Demo Day, July 1, 2026. Pitch and live demonstration of the 23 projects before a panel of healthcare and AI experts.

The programme in detail: what it makes possible, calendar, Demo Day, partners, questions

What the programme makes possible

Personalized care

AI agents that adapt to each patient’s context, drawing on the institution’s own clinical guidelines.

Care pathway coordination

Smoother handoffs between professionals, departments, and external partners across the patient journey.

Streamlined operations

Repetitive administrative work absorbed by agents, freeing teams for higher-value tasks.

Time reclaimed for caregivers

Less paperwork, fewer interruptions, more time at the bedside, where it matters most.

The programme calendar

Oct 2025 → Jan 2026

Open call

Each proposal is reviewed before selection.

H1 2026

Incubation

Projects are developed and tested in clinical settings.

1 July 2026

Demo Day

Public showcase of the projects built during the pilot phase.

2027

Extension

Roll-out to other public hospitals across France and Europe.

The Demo Day, in pictures

Highlights from our first Demo Day at Hôtel-Dieu, July 2026.

Partners

AP-HPHôtel-DieuBOPEXBOPA, Bloc Opératoire Augmenté (AP-HP, M2, Université Paris-Saclay)

Frequently asked questions

What does the program consist of?

A technology platform that lets hospital professionals develop AI prototypes — combined with methodology support, technical assistance, a secure regulatory framework, and open-source infrastructure.

How is data protected?

Data protection is an absolute priority. Every project complies with GDPR and the AI Act. Technologies are audited by independent third parties. Data is anonymized or pseudonymized, with regular security audits. No data is ever used for commercial or private research purposes.

Who takes part in the program?

The program is aimed at healthcare professionals working in a public hospital: physicians, paramedical and administrative staff, including residents with their supervisor's agreement. No prior AI or computing skills were required to join.

What kinds of projects does the program support?

AI projects that address concrete needs: organizing work (scheduling, repetitive tasks), care coordination, administrative processes, quality and safety. Early ideas as well as existing prototypes.

Is the program paid?

No. Impulse is a public-health innovation program dedicated exclusively to projects of general interest. The full program is free of charge for participating professionals and institutions.

How were projects selected?

Four criteria: relevance of the need (concrete, field-rooted problem), potential impact (improvement of care or working conditions), technical feasibility, and regulatory and ethical compliance.

What does the program ask of project leads?

Active participation. Project leads keep regular availability for collective workshops and working sessions, about two hours per week during the incubation phase.

impulse.healthcare ↗

Charité: complex data for health practitioners, in natural language

With the Institute of Medical Informatics at Charité, Berlin. The clinician opens a patient record, asks a clinical question in their own words, and the answer comes back with the measurements behind it. No query to write, no second tool to open.

From DotBase to Bayes Assistant, live.De DotBase à Bayes Assistant, en direct.

The clinician opens a patient in DotBase, clicks Open Bayes Assistant, and asks a clinical question. The connector queries InfluxDB and the chart renders in the conversation.Le clinicien ouvre un patient dans DotBase, clique sur Open Bayes Assistant, et pose une question clinique. Le connecteur interroge InfluxDB et le graphique s'affiche dans la conversation.
Charité
Institute of Medical InformaticsInstitute of Medical Informatics
Home›Beispiel, Ada›Test Health Metrics Visualization
Gekoppelt
AB
Beispiel, Ada
81 Jahre
22.06.2026Kliniker:inPsychiatrische Insti...
Test Health Metric [All Types]
Test Health Metrics Visualization
EingewilligtAktiv
17.06.2026Patient:inPsychiatrische Insti...
Test Health Metric [All Types]
Test Health Metrics Visualization
EingewilligtAktiv
30.04.2026Kliniker:inPsychiatrische Insti...
Test Health Metric [All Types]
Test Health Metrics Visualization
EingewilligtAktiv
Erfassungsdatum: 22.06.2026
Test Health Metrics Visualization
Heartrate
1401201008060
73.14
Apr 6Apr 13Apr 20Apr 27May 4May 11May 18May 25Jun 1Jun 8Jun 15
Blood Oxygen
10.960.920.88
1
Apr 6Apr 13Apr 20Apr 27May 4May 11May 18May 25Jun 1Jun 8Jun 15

Real InfluxDB connector, Ada Beispiel (pat-001), Apple Watch data, 27 Feb 2026Connecteur InfluxDB réel, Ada Beispiel (pat-001), données Apple Watch, 27 fév. 2026

Demonstration of the integration, on a sample patient.

Navigating a region’s care offer

12.6M inhabitants, with ARS Île-de-France and GIP SESAN

The ground

A care offer scattered across directories that do not talk to each other. The professional looking for the right service ends up making calls.

What the agent does

It rephrases, asks the questions that genuinely narrow the search, then shows the services on a map, nearby and in context.

What makes it usable

Several directories cross-checked, each with its link and its last update, so the professional knows what they are relying on.

A caveat the project itself carries: the agent helps with orientation, it does not replace medical expertise and is not a medical decision tool. Regional population: INSEE, 1 January 2026.

Employment

Alongside caseworkers and jobseekers

In 2016 we launched Bob with Pôle emploi: an open-source coach used by more than 500,000 jobseekers, one of the world’s first AI-powered public services.

Today the same Bayes Platform agents serve France Travail on both sides: in Neo, the caseworkers’ assistant, and in the jobseekers’ personal space.

500,000+

jobseekers supported by Bob

6M

jobseekers in France Travail’s scope

Neo: the caseworkers’ assistant, extended with specialist agents

Neo is the assistant France Travail’s caseworkers already open. The tool cards in its answers are Bayes Platform agents, called over MCP: the caseworker never changes screen.

NeoFrance Travail

Les outils que Neo appelleThe tools Neo calls

OutilToolCe qu’il interrogeWhat it queriesÉtatStatus
smart_searchRecherche assistée par IAAI-assisted searchOffres d’emploi, La Bonne Boîte, Mes Événements Emploi, data.inclusionJob offers, La Bonne Boîte, Mes Événements Emploi, data.inclusionBêtaBeta
get_job_offerDétail d’une offreOne offer in fullOffres d’emploiJob offersBêtaBeta
search_job_offersRecherche directe d’offresDirect offer searchOffres d’emploi, ROMEOJob offers, ROMEONon retenuNot integrated
Dossier individuelIndividual fileAucun outil n’y accèdeNo tool reaches itAucune API individuelleNo individual APIHors périmètreOut of scope
  • Appelé en MCP : notre serveur MCP expose ces outils, et Neo les appelle comme les siens. La ligne s’allume à chaque appel.Called over MCP: our MCP server exposes these tools, and Neo calls them as its own. The row lights up on every call.
  • Bêta : ouvert à un groupe de conseillers testeurs.Beta: open to a group of test caseworkers.
  • Non retenu : exposé par le serveur mais pas intégré par France Travail, donc Neo ne peut pas l’appeler.Not integrated: exposed by the server but not integrated by France Travail, so Neo cannot call it.
  • Hors périmètre : les sources sont des catalogues, jamais un dossier, et Neo le dit si un conseiller le lui demande.Out of scope: the sources are catalogues, never a file, and Neo says so if a caseworker asks.

Demonstration: an illustrative conversation on a fictitious search. The tools, their sources and their status are those of the real integration as of 24 September 2026.

The same agents become a personal coach for the jobseeker

CoachFT

A space reconstructed for illustration, and a demonstration scenario on test data. In service in the Gard since 7 September 2026, then rolled out department by department.

From the prompt that circulates to a governed agent

Prompts copiés-collés ou agents validésCopy-pasted prompts vs. validated agents

La demande venait du terrain et elle était juste. Ce que le fichier texte ne pouvait pas offrir n'est pas une meilleure réponse, c'est de savoir si elle est bonne.The need came from the field and it was a fair one. What the text file could not offer is not a better answer, it is knowing whether the answer is good.

prompt-ARE-v3-final(2).txtpartagé entre conseillersshared between caseworkers
Tu es un expert de l'assurance chômage à France Travail.

À partir des 24 derniers mois de salaire brut que je te
donne, calcule l'allocation d'aide au retour à l'emploi.

Prends le salaire journalier de référence, applique le
taux le plus favorable, puis vérifie les planchers et
le plafond en vigueur.

Tiens compte du motif de rupture, des périodes non
travaillées, et des indemnités de rupture qui décalent
le point de départ du versement.

Lis les pièces du dossier que je colle en dessous et
déduis-en ce qui manque.

Donne un montant mensuel et une durée, en euros, sans
réserve et sans me demander de précisions.You are an unemployment-insurance expert at France Travail.

From the last 24 months of gross salary I give you,
compute the return-to-work allowance.

Take the reference daily wage, apply whichever rate is
more favourable, then check the floors and the ceiling
in force.

Account for the reason for leaving, the non-worked
periods, and the severance pay that shifts the start
date of the payment.

Read the case documents I paste below and work out
what is missing.

Give a monthly amount and a duration, in euros, with no
caveats and without asking me for details.
  • Aucune gouvernance : ni qui l'a écrit, ni qui s'en sertNo governance: neither who wrote it, nor who uses it
  • Pas de version, pas de validation, aucune trace des réponsesNo version, no validation, no record of the answers
  • Le calcul est confié au modèle, sur un droit qui engage l'organismeThe calculation is left to the model, on an entitlement the organisation answers for

Extrait de prompt reconstitué et vue d'agent illustrative : c'est la forme de ces fichiers et de cette console, pas leur contenu exact.Reconstructed prompt extract and illustrative agent view: the shape of those files and of that console, not their exact content.

Social & Nonprofits

Helping nonprofits become agentic organizations

In 2024, CaseAI was the first AI copilot designed for social workers: notes became action plans, and profiles updated themselves.

Its features are being rebuilt on the Bayes Platform, so that every nonprofit can take them up and adapt them to its own work.

How we work with a nonprofit

01

A first use case

We start from one precise need of your teams, and build it with them on the platform.

02

Your teams design their agents

With no code, on your own rules, with validation before going live and monitoring after.

03

CaseAI’s features, rebuilt

Notes that become action plans, profiles kept up to date: they are being rebuilt on the platform, for every nonprofit.

You run a nonprofit. Let’s talk about your first use case. Talk to the team →

Governments

A state’s agentic foundation

Digital public infrastructure for agentic government

For a government, the Bayes Platform is digital public infrastructure: one shared, open-source and sovereign foundation on which every ministry builds its agents, instead of one tool per administration.

Just as mobile payments let countries leapfrog traditional banking, AI lets emerging countries leapfrog straight to AI-native government.

Citizen access

Public services in everyday channels

Every procedure in one conversation: messaging, local languages, voice notes.

Cross-ministry

One foundation, every ministry

Every ministry builds its agents on the same foundation, sovereign and open source, with the data kept in the country.

Back office

The case officer stays in control

The request lands in the officer’s own tool, where they see what the AI did, on which sources, and approve it.

Togo: an entire state, on one agentic foundation

Our pilot country: the national platform is being deployed with Agence Togo Digital, and the first citizen journey is going into service. Below, a diploma request handled end to end within the conversation.

République Togolaise

9M citizens in the state platform’s scope

An illustrative scenario, on the real journey going into service. Local languages and voice notes: including the roughly 40% of adults who can neither read nor write (CIA World Factbook, 2015).

And behind it, the back-office copilot

The citizen’s request lands in the case officer’s own tool, and they keep control: they see what the agent did, on which sources, and approve it. The specification the service wrote holds all the way into the back office.

Open the demonstration

Demonstration, on illustrative data.

Another sector?

We work with institutions, agencies and large nonprofits ready to co-build the AI infrastructure their sector needs.