Assistants that know your organisation
A conversational assistant grounded in your documents, procedures and vocabulary: M&E support, staff questions about internal procedures, help with drafting. Its answers rest on sources you can trace.
Custom AI tools · Humanitarian organisations
AI tools built for the field, and for your principles
I design specialised AI tools for humanitarian organisations: assistants that know your procedures, tools that produce your documents and analyses, connected to what your teams already use. Every tool follows the requirements I advocate as an adviser, and Opti', the platform I designed, is the proof.
Generic tools ignore your constraints
Consumer AI assistants can draft, summarise and translate. They know nothing of your procedures, your donors' vocabulary or the formats your teams produce every month. Above all, they were not designed for data about vulnerable people, for a connection that keeps dropping, or for teams working in four languages.
A specialised tool starts from the other end: from a specific task your teams do by hand today, and that takes them too long. AI goes where it adds something, and fixed rules everywhere else.
Sometimes the right answer is a tool that already exists. That is why a project can start with a benchmark of the available solutions, before deciding to build.
02 · What I build
Specialised assistants
Production
Integrations
Controlled data
Four kinds of tools, depending on your needs
Documents and analyses
Terms of reference, questionnaires, analysis tables, reports: tools that produce your deliverables in your formats. Figures are calculated by code, never by the language model, and every document is still reviewed before release.
Connected to your tools
Your teams already work in KoboToolbox, Excel or web tools. A useful tool plugs into them, instead of asking people to change how they work.
Hosting under control
For the most sensitive data, AI hosted on dedicated infrastructure that goes through no commercial provider. I did this for Opti' with OptIA, and we can look together at whether that choice makes sense for you.
The requirements behind every tool
I design to the five TRUST requirements, my reading of the SAFE AI sector framework. Applied to building a tool, they become concrete technical choices:
T
TransparencyEvery processing step is readable. Users see what the tool is about to do before it does it.
R
ReliabilityCalculations are done by code. The model interprets and writes from verified results.
U
UsefulnessAI steps in where a fixed rule is not enough: understanding an open answer, structuring a document, summarising an interview.
S
SupervisionEverything produced stays editable, and nothing goes out without human approval.
T
TightnessYou know where the data goes, it is encrypted, and it trains no model.
Field constraints are built in from the start: unreliable connectivity, hence offline operation where needed; multilingual teams; modest devices, often a phone.
Opti': the approach in production
Opti' is the monitoring and evaluation platform I designed and run, used by more than 800 professionals in over ten countries.
Opti' covers an M&E assignment from end to end, from terms of reference to the final report, with a specialised assistant, a project workspace and a training coach. It is built to the requirements described above, which is what makes it a good example of what I can build for an organisation.
- Figures that are not made upIn generated reports, the values quoted come from a register calculated from the data. The writing only references them.
- A check before the writingBefore writing a report, the tool shows the statements it is about to make and the figures behind them. The user approves, corrects or rejects each point, and adds the context only the field knows.
- Data collection that works without a networkQuestionnaires are filled in offline, and sync when the connection returns.
- Protected dataSensitive content is encrypted in the database, with a key kept outside the database. For the most sensitive cases, OptIA processes requests on dedicated infrastructure, with no storage and no third-party provider.
“What I built for Opti', I can build for an organisation, with its own data, procedures and teams.”
The platform is described in full on myoptibot.com.
How we get started
- A first 30-minute callFree and with no commitment. You describe the work you want to equip, the data involved and the people who will use the tool.
- A written proposal within 48 hoursWhat the tool will do, what it will not do, and how we will build it.
- Building it with the teamsThe tool is tested on your real cases, by the people who will use it, before it is rolled out.
If the first question is which tools to allow, and under what rules, that is the purpose ofAI advisory.
A task to equip?
Describe it to me in a few lines. I will tell you what AI can bring to it, and what it should not do there.