Founder
A mathematician-economist, data engineer and AI researcher, I have spent fifteen years building decision systems for Malagasy public institutions and organisations. I founded Cabinet IDEA and co-founded the Idea Academy NGO so that these tools would not remain the preserve of a connected minority.
One simple conviction: a technology that only reaches those who already have access to everything develops nothing. It widens the gap.
Four convictions, formed in the field rather than in a seminar room. They explain why my projects rarely look like what is usually proposed in Madagascar.
Accessibility
At most a fifth of the population has internet access. A platform built for the smartphone excludes most of the country from the outset. I start from the poorest channel — voice, SMS, USSD, radio, field agent — and build upward. The reverse never works: the low-bandwidth version promised “for later” never arrives.
Democratisation
A platform delivered without local skills dies when the funding ends. Every assignment I run includes the handover: internal teams must be able to evolve the tool without me. That is the reason the Idea Academy NGO exists, and the yardstick I accept being judged by.
Sovereignty
Malagasy public data, health data and agricultural data belong to those who produce them. I design architectures where hosting, models and access rules stay under local control — and where any data collected from a community comes back to it in a form it can use.
Pooling
Health, education, economy, governance, environment: five domains, but one citizen, one field agent, one channel. Funding the same telecom gateway five times means funding inclusion nowhere. That question became my doctoral research subject.
Public finance, university teaching and data engineering — three trades that meet at one point: understanding public decisions, knowing how to teach them, and knowing how to equip them.
Janv. 2026
Directorate General of the Treasury — training centre
Intensive hands-on sessions on artificial intelligence and data analysis, designed for civil servants with no prior technical background.
+50 officers
2024 — aujourd’hui
Idea Academy NGO
Instructional design and rollout of the Data & AI programmes across the regional universities of Diego, Fianarantsoa and Toliara. Funded by YAS Madagascar.
+700 beneficiaries
Juil. 2024
Directorate General of the Treasury
Technical design and deployment of a data platform, followed by training internal teams on decision-support tools.
15 officers trained
2023
University of Antsiranana, Economics department
Theoretical and applied econometrics, taught entirely through Python programming.
+50 students
2021 — aujourd’hui
Cabinet IDEA and Idea Academy NGO
Training the trainers, designing and deploying the e-learning platform, and applied research (NeuralProphet, interbank machine learning).
+20 trainers
2020 — aujourd’hui
Cabinet IDEA (Intelligence Data Exploiting Agency)
Predictive modelling, decision dashboards and machine learning pipelines for private and public clients, with skills transfer to the client teams.
Nov. 2018 — aujourd’hui
Directorate of Studies, Ministry of Economy and Finance
Analysis of the impact of public finances on the national economy, technical liaison with the IMF, the World Bank and the AfDB, and internal capacity building.
2018 — 2020
GREFTP V7V Manakara — technical and vocational training
Financial, accounting and administrative management of the institution, including financial monitoring of its training programmes.
2016 — 2022
University of Toamasina
Academic teaching in economics and finance, from bachelor to master level.
+70 students
2015 — 2018
General Treasury of Manakara
Oversight of regional public accounting and continuing education of local accounting officers.
2015
Malagasy Institute of Planning Techniques
Thesis: medium-term projection of the macroeconomic and social variables of the Malagasy economy under an economic model.
2012
University of Antananarivo
Thesis: inflation dynamics, exchange rates and the new open-economy macroeconomics.
2012
University of Antananarivo, Economics department
Econometric modelling of time series for the finance postgraduate programme, and measure theory applied to economic analysis.
+120 students
2010
University of Antananarivo
Thesis: modelling the equilibrium real exchange rate — the case of Madagascar.
2009
University of Antananarivo
Thesis: affine codes and antichains.
Artificial intelligence, LLMs and agents
ExpertGoverned assistants, document RAG and autonomous agents wired to business data — in production, not in demo.
Open WebUI · MCP · Mastra · vLLM · pgvector · RAG
Machine learning and data science
ExpertMacroeconomic forecasting, time series and risk models, from prototype to an industrialised life cycle.
NeuralProphet · PyCaret · MLflow · scikit-learn · Pandas
Data architecture and reporting
ExpertEnd-to-end data platforms: ingestion, warehouse, semantic layer, reporting and access governance.
Airbyte · Superset · Directus · Next.js · PostgreSQL · MinIO
Decision-support application development
ExpertDecision-support applications and data pipelines built to be taken over by the internal teams.
Python · FastAPI · SQLAlchemy · Pydantic · Streamlit
Econometrics and macroeconomic modelling
ExpertStructural analysis, forecasting and assessment of how public finances affect the national economy.
Séries temporelles · Modèles structurels · Finances publiques
DevOps and MLOps
AvancéGetting platforms and models into production and keeping them running, under real infrastructure constraints.
Docker Swarm · ArgoCD · Harbor · GitLab CI/CD · Helm
Financial modelling
ExpertModelling and analysis of complex financial data, down to the tools management actually uses.
Excel · Power Query · Power Pivot · DAX
These platforms are not mock-ups. One serves a national administration today; the others are being built to the same standards of accessibility and governance.
Governance, public finance and transparency
Decide on figures you can trace back to their source.
Public decision-support and accountability system. It consolidates governance and public finance data, makes it traceable back to its origin, and opens a citizen grievance channel reachable from a basic handset.
Documented capabilities
Public institutions · Court of Auditors and oversight bodies · Civil society and media · Local authorities
See the platformEducation, teaching and training
Train, employ, reintegrate — in the age of artificial intelligence.
National adaptive learning system, from basic education to continuing professional training. It carries learning paths, employment and reintegration, civic education, and reaches families who have no Internet access.
Documented capabilities
Ministry of Education and inspectors · Schools and teachers · Young people, learners and families · Companies and training funders
See the platformEconomy: farming, livestock, fisheries, tourism
Market prices and the right farming decision, down to the village.
Rural economy support system: commodity prices, crop diagnosis, weather alerts and input pooling. It is built first for the isolated producer, through the channel they already own.
Documented capabilities
Producers and cooperatives · Regional agriculture and livestock directorates · Exporters and value chains · Rural development programmes
See the platformPublic health and primary care
Medical decisions at the last mile, in thirty seconds.
Community health support system. It equips community health workers, puts triage within reach of a free call, and hands the doctor a prepared decision to validate in one click.
Documented capabilities
Ministry of Health and health districts · Basic health centres and chief doctors · Community health workers · Health and research partners
See the platformPersonal safety, property protection and environment
Protect people, property and forests — with the community, under legal authority.
Integrated protection system in three strands: people, property and environment. No response is ever triggered by the AI: only a judicial officer can qualify an alert and authorise action.
Documented capabilities
National gendarmerie and judicial authorities · National disaster management office · Forestry service and protected area managers · Municipalities and community committees
See the platformDoctoral research
Where the digital divide is extreme, inclusion is not a channel problem — that one is solved and documented. It is a problem of architecture and governance: the universal channel has a high fixed cost and a low marginal cost, while public funding is carved up by sector. Each project therefore has to bear alone a cost it cannot carry, and falls back on the already-connected minority.
Q1
Which building blocks of a cross-sector AI system must be pooled, and which must remain sector-specific?
An operational criterion, usable before any spending is committed, by architects and funders alike.
Q2
Does the marginal cost of a service for offline users actually fall when the foundation is pooled?
A cost-per-person-actually-reached measure, usable as a public decision-making instrument.
Q3
How do you measure the inclusion an AI system actually produces, beyond sign-ups and message volume?
An indicator centred on the decision the user changed, broken down by level of digital maturity.
Q4
How do you govern data when the person concerned has no access to administrative writing, no connection and no personal device?
A framework for spoken consent, withdrawal over universal channels, and giving data back to the community that produced it.
Q5
Which funding, governance and organisational forces produce the silo, and what contains it?
A longitudinal account of four real platforms, measured as it happened rather than reconstructed afterwards.
Q6
Does the value delivered to offline users depend more on the field-agent network than on model performance?
A documented split of intelligence between the model, the field agent and the user.
Public institution, funder, university or company: if your project depends on data and AI actually reaching the field, let us look at the problem together.