ConfluxLab

An independent research laboratory

Applied intelligence at the confluence of finance and humanitarian practice.

ConfluxLab studies how computational methods shape high-stakes decisions — in capital markets and in humanitarian operations — and publishes the methods, evidence, and critique that those decisions need.

Explore our research → Partner with us →

About the laboratory

An independent laboratory at a methodological divide.

Conflux Laboratory — ConfluxLab for short — is an independent research laboratory working at the confluence of computer science, quantitative economics, and humanitarian practice. We produce peer-reviewed research, open methodological tools, and applied collaborations with financial institutions, regulators, humanitarian organisations, and donors.

The lab exists because the most consequential uses of machine learning now occur in sectors that rarely share methods with one another. Capital markets have spent decades refining statistical modelling under adversarial conditions; humanitarian agencies have spent decades refining operational judgement under resource constraints. Neither set of hard-won knowledge travels well across the divide. ConfluxLab was founded to narrow that distance — not as a metaphor, but as a concrete research programme.

We work in the open by default. Methods and working papers are published without paywalls; code and — where ethically permissible — data are released alongside them. Where the work involves vulnerable populations, we treat data protection and methodological rigour as inseparable requirements and decline projects that cannot meet both.

Research

Three pillars, one methodological vocabulary.

Each pillar publishes independently; each informs the others.

Quantitative Finance & RegTech

Models the people who sign decisions off can use.

We work on applied machine learning and statistical modelling for financial markets, credit and operational risk, and the automation of regulatory workflows. Current threads include interpretable risk models for supervisory use, the analysis of unstructured regulatory and disclosure text, and evaluation methodology for algorithmic trading systems.

We collaborate with quantitative research teams at banks, asset managers, and supervisory bodies. Work in this pillar is designed to be usable by the people who have to sign decisions off — not only by the people who build the models.

AI for Humanitarian Action

Computational methods, run with agencies, not around them.

We develop and evaluate computational methods for humanitarian operations: early-warning systems for displacement and food insecurity, forecasting of needs across protracted crises, operational analytics for NGOs, and responsible data practices for work with affected populations.

Projects are run with humanitarian agencies on the ground, not around them. A portion of the lab's capacity is reserved each year for pro-bono engagements with organisations that could not otherwise commission this class of research.

See work in Humanitarian Action →
AI Governance & Societal Impact

The same questions, answered once across both sectors.

A cross-cutting programme on algorithmic audit, ethics and risk review of deployed AI systems, and policy research on the regulation of AI in finance and in humanitarian settings.

The questions raised by models in one sector — fairness, interpretability, liability, data provenance — are usually the same questions raised in the other. This pillar exists so they are answered together, not twice.

Case studies

Applied collaborations, documented in the open.

Two of the lab's humanitarian engagements — one in pilot, one in its research and data-collection phase. Each is built to produce a public artefact alongside whatever the partner needs.

Pilot · Zimbabwe · with EthicalChain AI for Humanitarian Action

Detecting child labour in agricultural supply chains

A computer-vision monitoring method that supports the Child Labour Monitoring and Remediation System (CLMRS), now in pilot with ethicalchain.org. The work studies whether edge-deployed detection can make supply-chain monitoring more systematic without moving sensitive imagery off-site.

Read the case study →
Research & data collection AI for Humanitarian Action

Allot — a moral laboratory for humanitarian allocation

A cross-cultural study of how people allocate scarce aid under genuine scarcity. The curated, openly published signal is intended to inform a later allocation-support system, developed with omniaid.io — trained on human moral judgement rather than cold optimisation.

Read the case study →

How we work

Method before demonstration.

  • Applied, not demonstrative. Our research questions are chosen because the answers change decisions. We do not publish demonstrations; we publish methods, evaluations, and critique that institutions can act on.
  • Open by default. Working papers, methods, and — where ethics permits — datasets and code are released openly. Closed publication is the exception and is justified in writing.
  • Partnership over consulting. We work with partners as co-investigators, not vendors. Every engagement produces a public artefact (a working paper, an open tool, a methods note) alongside whatever internal deliverables the partner needs.
  • Humanitarian pro-bono track. Each year we reserve a defined share of lab capacity for humanitarian organisations that cannot afford commissioned research. Selection criteria and past projects are published.
  • Ethics is not a gate, it is part of the design. Data protection, informed consent, and risk review are embedded in project design, not applied as a final check before release.

Ethics

Ethical review is part of research design, not a step at the end.

ConfluxLab treats ethical review as part of research design, not as a compliance step attached at the end. Every project the lab accepts is evaluated against three questions:

Whose decisions will this model change? Whose data is being used, and on what basis? What happens if the model is wrong?

Projects that cannot produce adequate answers are not accepted. Work involving vulnerable populations follows the principles laid out by the humanitarian community on responsible data — minimisation, informed consent where possible, non-identifiability, secure storage, defined retention, and honest communication of uncertainty. Work involving regulated financial activity follows the applicable rules of the jurisdictions concerned and documents the chain of responsibility between model, output, and decision-maker.

The lab will decline, and has declined, engagements where the expected use of a model conflicts with these principles. We publish a short annual note describing the shape of that refusal rate without disclosing partner identities.

Contact

Considering a collaboration?

We work with partners as co-investigators. Choose the relevant address below — it opens in your own mail client.

General enquiries
Research collaboration
Partnerships
Media
LinkedIn
linkedin.com/company/confluxlab

Email addresses open in your own mail client when you select one. They are not published in plain text, to limit automated harvesting — replace [at] with @ and [dot] with . if you prefer to type one manually.

A monthly note on what the lab has published, what we are reading, and open calls. No tracking beyond what is required to send the email.