Technology that turns data and real-world problems into working systems.

Four practices, one way of working.

We build technology around a problem someone can describe in a sentence. If we cannot find that sentence, we do not start.

Three questions. One honest starting point.

No form, no email gate. The answer tells you which of our practices fits — including when the answer is “you do not need us yet”.

Start with Agrima

A management platform for animal farming — feed, health, finance and daily operations in one record. We map your existing routine before anything is installed.

Start with a data and AI review

Before any model, we look at what you already collect and what decision it should support. Most engagements start with a two-week review, not a build.

Start with digitization

We digitize the records that already run your business — stock, customers, staff, finance — so nothing depends on one notebook or one person.

Start with a custom platform

When nothing off the shelf fits, we design around the workflow you actually use. You get the architecture notes and the source, not a black box.

Set your flock size, feed price and current conversion ratio, and see what a 0.1 improvement is worth per cycle. The formula is printed under the result — no hidden multiplier.

AI that speaks the language the work happens in.

What sustainability and transparency mean here.

Sustainability is an engineering constraint here, not a slogan — and transparency means you can check our work.

Five steps, and you can stop after any of them.

Each phase produces something you can use on its own — a scope, a prototype, a trained team.

The things people ask on the first call.

Have a problem worth solving?

Tell us what is not working. If technology is the wrong answer, we will say so on the first call.

Agriculture technology

Platforms that monitor feed, health, inputs and yield so a farm runs on evidence instead of memory.

AI & machine learning

Language models, forecasting and decision support built for local context — including Kinyarwanda.

Digital platforms

Websites, portals, mobile apps and management systems designed around how your team already works.

Consulting

Short, scoped reviews that tell you what to build, what to buy and what to leave alone.

You can see the method

Every engagement ships with the architecture notes, the assumptions behind each number, and the limits of what the system can tell you.

You keep the keys

Your data stays in infrastructure you control, and the source we write for you is yours. No lock-in clause, no hostage migration fee.

Built to last, not to churn

We design for low bandwidth, intermittent power and offline days — so a system keeps working on the worst week, not just the demo.

Two weeks on site and in your records, before a line of code.

On site, in the records, with the people doing the work. We write down the baseline.

A written scope with what is in, what is out and what it costs.

A scope naming what is in, what is out and what it costs. You could take it elsewhere.

Short increments you can see and challenge every fortnight.

Fortnightly increments you can use and criticise, not a reveal at the end.

Training in the language of the people who will use it daily.

Rollout with training in the working language, and a fallback for the first weeks.

Measured against the baseline we agreed at the start.

Reviewed against the baseline. If it did not move, we say that too.

Agrima

Agrima connects commercial livestock operations to high-fidelity telemetry, dynamic feed conversion modelling, health surveillance and audited margin intelligence — engineered for rugged barn conditions first and the executive dashboard second.

In-barn telemetry

Load cells under the silo, hopper weight logging, auger run-time and microclimate probes for temperature, CO₂ and humidity.

Edge diagnostics

Optical sensors compute bird distribution and huddling; acoustic arrays flag respiratory stress before it is audible to a person.

Predictive FCR engine

Daily intake modelled against weight gain to project the ideal harvest window and avoid feeding past the economic optimum.

Unit economics

Cost per kilogram of liveweight gain computed daily, with batch records exportable for buyers, lenders and auditors.

Feed optimization & dynamic FCR

Feed is up to 70% of lifecycle expenditure. Agrima reconciles recorded feed drops against silo weight, so clumping, moisture and ration variance surface in days rather than at settlement.

Flock & herd health surveillance

Mortality and morbidity clustering detects sub-clinical enteric or respiratory challenge ahead of an acute spike. Digital necropsy logging pairs barn managers with remote pathologists in Kigali and Montreal.

Feed is 70% of the cost base

Silo weight and intake reconcile daily, so drift is visible the same week.

Health signals arrive early

Clustering on mortality and behaviour flags a problem before it becomes a spike.

Works with the network down

The barn node records locally and syncs when the link returns.

Umuseke

General models stumble on agglutinative, dialect-rich Bantu languages — they fragment the words and invent the context. Umuseke is a language engine trained for Kinyarwanda morphology, grounded in local agricultural and public-sector sources, and packaged to run where your data already is.

Acoustic & dialect ingestion

Speech captured in the field is normalised for pitch accent and regional variation before anything else happens.

Morpho-syntactic engine

Words are decomposed into their grammatical parts, so meaning survives the agglutination that breaks generic models.

Grounded generation

Answers are drawn from a vetted corpus, with the source retrievable rather than implied.

Sovereign dispatch

Results return over the channel that suits the user — app, SMS, IVR or API — without leaving your infrastructure.

A byte-pair tokenizer calibrated on 16 Bantu noun classes and infix grammar, so a word survives tokenization intact.

One Kinyarwanda word can carry what English spreads over a clause. General tokenizers shatter it into fragments that mean nothing.

Retrieval grounded in agricultural extension handbooks and national gazettes, with a citation attached to every answer.

Frontier models improvise agricultural advice for a context they have never seen, confidently and incorrectly.

On-premises appliances in Kigali or Montreal, and a fully offline mode for field devices.

Sending citizen or farm records to a third-country API is often not legal, and rarely wise.

4-bit quantized weights running on rugged quad-core field tablets, with no network dependency.

A round trip to a distant data centre is not viable on an intermittent rural connection.

Agronomic voice advisory

Farmers ask a question by voice in Kinyarwanda and receive a grounded answer. Works over IVR and messaging gateways, so no smartphone is required.

Government & digital public infrastructure

Translation and document understanding for public services, deployed inside the institution rather than called out to a foreign API.

Financial & insurance verification

Claims and loan documentation read in the language they were written in, cutting the manual translation step out of the approval path.

Sovereign knowledge bases

Internal documentation made searchable across Kinyarwanda, French and English, hosted in private clusters in Montreal and Kigali.

Runs where the data is

On-premises or fully offline, so field recordings never leave the country to be understood.

Tokenized for Bantu morphology

A word survives tokenization intact instead of shattering into meaningless fragments.

Grounded, with citations

Answers are drawn from a vetted corpus and carry the source they came from.