Huzalabs knows AI needs more than code

Thirty-five women entered an AI and data-science bootcamp in Kigali. Thirty-five graduated. In a field fond of enormous numbers, this small one deserves attention. It describes a complete cohort, a finite programme, and a company doing something more patient than announcing that the future has arrived. Huzalabs, which delivered the training, also builds the software that such a future would require.

• The work: custom software, AI consulting and digital skills training.
• The buyers: businesses, governments and NGOs with systems to build or services to digitise.
• The distinctive combination: engineering, locally relevant language work and practical help competing for contracts.

The classroom changed the brief

The March-August 2024 programme ran three in-person sessions a week. There were 1,413 applicants, 100 shortlisted candidates and 35 places. Originally intended for women with little IT experience, it shifted toward women with IT backgrounds after organisers examined the applications and employment challenges. Transport and data stipends helped participation.

That change is the interesting part. A curriculum can be beautifully written and still address the wrong starting point. Here, the applicants supplied information that altered the programme. For anyone organising technical training, the implication is straightforward: investigate who needs the course before becoming excessively attached to the course you imagined.

“Participants were more open to sharing their ideas, seeking clarification and actively contributing to discussions without fear of judgement.”Huza Labs, Rwanda / FAIR Forward impact report

Huzalabs’s observation about women-only settings gives the graduation figure some texture. Participation has a social architecture as well as a syllabus. A person who can ask a question has a better chance of resolving it. The practical lesson is to treat learning conditions as part of the technical design.

Software with an institutional brief

The company’s commercial work sits alongside this educational activity. Huzalabs describes its customers as SMEs, corporations, governments and NGOs. Its expertise spans natural language processing, computer vision, big-data engineering, IoT and scalable software. Those capabilities are ingredients; the buyer’s problem determines the recipe.

Its advertised solution areas include digital identity, electronic notarisation, healthcare platforms, contact centres, education systems and energy and climate intelligence. These are places where software must interact with established procedures. An identity system, for example, has to establish trust across services. A health platform must make information useful to people whose decisions cannot wait for a charming demonstration.

The service offering includes digitisation and cloud migration. Read together, the catalogue suggests a company working across the layers of organisational change: infrastructure, applications and the people operating them. That is an interpretation of its portfolio, rather than a claim that every advertised system has been deployed.

For a customer, the attraction is the possibility of discussing a working process as well as a codebase. The trade-off is scope. A broad consultancy brief needs clear deliverables, ownership and acceptance criteria. Otherwise, a conversation about transformation can become wonderfully expansive while the original operational problem remains at its desk.

Kinyarwanda belongs in the specification

Huzalabs’s public GitHub repositories provide a more particular view of its interests. They include Kinyarwanda speech-to-text model assessment, government-web translation, mathematical-lesson translation and multilingual feedback analysis. Repository names indicate areas of work; they do not, by themselves, prove a production deployment or measured accuracy.

Still, the pattern matters. Local language is part of the engineering agenda. A system that handles an organisation’s information must also handle the language in which that information arrives. Translation quality and speech recognition can determine whether a supposedly convenient interface is convenient for its intended users.

A December 2022 participant account describes an NLP fellowship introduced by Huzalabs and Localis with GIZ, the Digital Transformation Center Rwanda and FAIR Forward support. The early work involved extraction, cleaning, preprocessing and predictive models. Anyone expecting a chatbot to emerge fully dressed from a prompt might find the laundry instructive.

The research connection extends further. Founder and CEO Davy K. Uwizera co-authored the 2023 Nature Human Behaviour article A new research agenda for African generative AI. Its concern with locally relevant development fits the practical language work. Context becomes something to investigate, test and build around.

The paperwork before the contract

Huzalink addresses a different bottleneck: competing for business. The platform offers opportunity intelligence, compliance support, proposal preparation, partner matching and expert sourcing. Its audience includes businesses and freelancers pursuing contracts. The promise is assistance through the bid process, with technical and professional resources brought into the same place.

A useful way to understand that offer is to follow a hypothetical bidder. Finding an opportunity is one task. Demonstrating compliance, assembling suitable expertise and preparing a credible proposal are several more. Huzalink groups those activities. That grouping explains its place in Huzalabs’s portfolio more clearly than a general promise about business growth.

Two other products broaden the picture. Huzaflix is marketed for generative-AI software building with front-end, back-end and API integration. Huzanetics is presented as mobility-data intelligence for emergency response. These are distinct propositions, and a prospective customer should assess each against its own requirements.

For buyers comparing alternatives, Huzalabs sits between a custom development agency, an AI consultancy and a training provider. An internal team could also handle parts of the work. The useful comparison is whether the chosen arrangement covers the whole assignment: building, integration, staff capability and continued operation.

Start with the decision you need to change

The business combines institutional software services, consulting, training and proprietary products. Rather than begin with a shopping list of AI technologies, a buyer could bring Huzalabs one recurring problem: an overloaded service channel, an inaccessible information flow, or a bid process missing essential expertise.

Then define what improvement would look like. Which users benefit? What data is available? Who maintains the system? What will count as successful delivery? The classroom offers a useful precedent for that conversation: examine the actual starting conditions, change the brief when the evidence warrants it, and make the work specific enough to finish.