Expert human data for AI teams, delivered from Africa. Language specialists and credentialed professionals evaluate models, build datasets and test AI assistants, and are paid fairly to do it well.
“Ninawezaje kutuma pesa kwa simu?”
“How do I send money by phone?”
Built for teams making AI work in the real world.
For AI labs, data vendors and AI product teams. We turn expert human judgment into data you can audit: every project comes with accuracy, agreement and turnaround reporting.
Find where your model gets it right and where it fails. Evaluators score answers against a written rubric and test safety in the languages your users speak.
Native speakers who know the idioms, code-switching and everyday context that global models miss, for text and speech.
Credentialed African professionals write and grade answers in fields where accuracy matters, with registration checked before their first task.
Need the fundamentals too? Document extraction, classification and image annotation are available as add-ons to the services above.
↗For product, design and research teams. We recruit verified participants for interviews, usability tests and surveys, handle screening, scheduling and incentives, and replace no-shows. You pay per participant, with no subscription.
Per participant, general consumer audiences, incentive included. Local-language moderation, transcription and synthesis available as add-ons.
For banks, insurers, telcos and fintechs deploying AI assistants across Africa. African experts build a test suite around your products, your customers’ languages and local rules. You own the suite and can rerun it on any model.
How each task is built
Test categories
We publish what we learn about model performance on African languages and professional tasks, and license the datasets we build.
A public benchmark of frontier models on African tasks: Swahili and Sheng customer support, tax and accounting, health, law and education. Written by practising professionals, graded by rubric.
About the IndexLicensable African-language and African-context datasets, from Swahili and Sheng intent data to expert-graded finance questions. Non-exclusive licences, with samples on request.
See planned datasetsWe start with East Africa. Each language has a lead who writes its guidelines and settles disputed labels.
Language scope and specialist availability are agreed for each project.
A clear rubric, the right people and checks at every stage. Pilots start with an accuracy target agreed up front.
Turn your goal into a written guideline, a rubric and a gold set of answered examples.
Model-assisted triage clears easy items so experts spend their time on hard ones.
Vetted language specialists and credentialed experts do the judgment work.
Multi-pass review, gold-set accuracy, agreement checks and weekly calibration.
Secure hand-off with accuracy, agreement and turnaround reporting.
Behind every label is a person. Core workers are employed, not gig contractors, and pay is published before anyone applies.
Rates for each role are public, set above local benchmarks for AI data work, and never cut while a project runs.
Counselling access, a grievance route with independent escalation, and no moderation of graphic or abusive content.
Managed devices, browser-only data access, project-level permissions and audit logs. We are working toward SOC 2; certification is not yet held.
Fasili.ai is being built to close a simple gap: AI teams need trusted human judgment for African markets, and talented people across Africa deserve well-paid, well-run work providing it.
Fasili is Swahili for a definition or an interpretation, which is what our people give your data.
Fasili brings together data science, African context and professional expertise. Data scientists design the rubrics, gold sets and quality checks behind every project. Language specialists bring the idioms, code-switching and everyday references that global models miss. Credentialed accountants, clinicians, lawyers and teachers judge whether an answer is actually right. The same verified people also take part in user research, so AI and product teams can test their products with the people who will use them. Every project reports accuracy, agreement and turnaround, so clients can check the quality of what they receive.
Tell us what you’re working on. Start with a focused pilot, then build a partnership around what your team needs.
