Fine-tuned & on-device models
Custom language modelsModels fine-tuned on your domain and compressed to run on the hardware you have: a phone, a laptop or your own server. No per-query fees to a foreign API, and it keeps working offline.
AI development · Banjul, The Gambia
AI development
AI development
in The Gambia.
HexAI designs, trains and ships AI for Gambian institutions and businesses: fine-tuned models, on-device inference, AI automation and custom machine learning, built in Banjul and running in production.
What we build
AI that does a real job
We don't wrap someone else's chatbot and call it a product. We start from the problem, pick the smallest model that solves it, and engineer it to run where your users are, including on budget phones with no signal.
AI automation for businesses
Workflows & document processingAutomating the repetitive work that eats staff time: reading receipts and documents with OCR, routing requests, drafting replies, and flagging the cases a person needs to see.
Custom machine learning
Prediction & decision supportMachine learning models built on your own data for forecasting, classification and recommendations, with evaluation you can audit before anything goes live.
Proof · AI we've shipped
Built and running today
Bantaba AI · On-device modelLive
An AI tutor that runs with no internet Fine-tuned to teach through questions.
We fine-tuned a 0.6B-parameter model at the weight level to teach data science and machine learning the Socratic way: it guides with questions instead of handing over answers. It is quantised and compiled to run fully offline on budget Android phones, with automatic switching to an online mode when a connection is available.
0.6B
Parameters, fine-tuned58MB
App size5–7s
Offline reply on a 4GB phoneAdaptive AI maths tutor
Personalised learningModes for kids, teens and exam-takers, with lessons and examples localised to five African countries and pace that adapts to each learner.
karang.hx-ai.org ↗Offline exam prep with AI explanations
WAEC · GABECE · JAMB20,000+ past questions that work offline, with Socratic AI explanations that show students how to solve a problem, not just the answer.
base10.gm ↗
How an AI project runs
From idea to something people use
Scope
We define the problem, the users and what success looks like, and check whether AI is the right tool at all.
Data
We gather, clean and structure the data the model needs, including digitising paper records where necessary.
Build & evaluate
We train or fine-tune, then test against agreed benchmarks before anything reaches a real user.
Ship & maintain
We deploy where your users are, monitor it in production and keep improving it after launch.
Common questions
AI development FAQ
Does HexAI build AI in The Gambia, or outsource it?
We build it ourselves. HexAI's engineering team is based in Banjul and has fine-tuned, deployed and maintains the AI products shown on this page.
Can AI work for users with poor or no internet?
Yes. Bantaba AI runs a fine-tuned model entirely on a budget Android phone, and Base10 works offline and syncs when students reconnect. We design for the connectivity your users actually have.
Do we need a lot of data to start?
Not always. Many projects start from an existing open model fine-tuned on a modest amount of your data. During scoping we tell you honestly what data is needed and whether it is available.
How much does an AI project cost?
It depends on scope, so we quote per project after a free 30-minute strategy call. We'll recommend the simplest approach that solves the problem, which is often cheaper than you expect.
Start a project
Have a problem AI could solve?
Tell us about it. We'll tell you honestly whether AI is the right tool, and what it would take.