Evidence first
No text is produced without a citation. Everything traces back to its source.
About
Assessment deserves better than an opinion.
Themiros grew out of a simple observation: in assessment, conclusions travel faster than the evidence behind them.
The problem
Assessing a policy, programme or project is slow, costly, and often hard to defend. The corpora are large, heterogeneous, multilingual. Manual analysis takes weeks. And at the end, a score built on a handful of sentences does not hold up in front of a funder who asks what it rests on.
The deeper problem lies elsewhere: most assessments blend what is measurable with what is a matter of judgement. Once the two are conflated, fact can no longer be told apart from appraisal — nor either be discussed without calling the other into question.
Our stance
We are not trying to produce the best score. We are trying to produce a score whose origin can be explained. The value of an assessment does not come from the number: it comes from being able to trace back to the sentence that justifies it, and from knowing where the tool could not conclude.
A tool that refuses to conclude when the corpus is insufficient is more useful than a tool that always concludes.
What we do not do
We do not produce a single overall grade: it conceals more than it informs. We do not do causal inference — Themiros assesses, it does not explain why. We do not display sentence-by-sentence confidence, because confidence is built from the convergence of a corpus, never from an isolated excerpt. And we do not replace the evaluator: final approval is theirs.
Our principles
Six rules, non-negotiable.
They are written into the product, not just onto this page: each one maps to a technical safeguard.
Visible rules
Objectified scores rest on auditable weightings.
The right not to conclude
Insufficient corpus: “unknown” or “to document”, never a forced conclusion.
Confidence from the corpus
It is built from documentary convergence, not from one sentence.
Multilingual by default
A document is never ignored because of its language.
The human decides
Final approval is human, and it is traced.
Who we work with
Those who have to answer for it.
Governments, national agencies, funders and funds — organisations for whom an assessment carries commitment.
Ministries and agencies
Preparing an NDC, reporting on it, or deciding between programmes.
Funders and funds
Reviewing applications and having to justify their funding decisions.
Evaluation firms
Producing reports that must stand up to challenge.
The team
The people building Themiros.
A small team, straddling assessment and engineering.
Themiros was built on a shared conviction: assessment does not suffer from a lack of intelligence, it suffers from a lack of traceability. The product grew out of the meeting between those who run these assessments in the field and those who know how to turn a method into a verifiable system.
Two sides, one requirement. Domain expertise defines what deserves to be measured and what is a matter of judgement — the boundary most assessments blur. Product and AI engineering holds that boundary inside the tool: every score stays tied to the sentence that grounds it, whatever the language of the document.
Pauline Coquerelle
CEO & Climate Resilience Expert
Antsa Razafimbelo
CEO & Lead Sustainable Development Expert
Meva Soavina Ravelo
Lead Tech & Needs-to-Product Expert
Ny Avo Volasoa Rakotoniaina
AI Developer & NLP Expert
Let us talk about your next assessment.
Thirty minutes is enough to see whether the method fits what you need to produce.