Kitovu: the cord that carries life to where it is needed

We help make your AI ideas real.

TovuTech is a Kenyan AI company working end to end. We define the workflow with you, design the data and machine learning pipeline, build and deploy the solution, then analyse the results and publish them as a notebook or web page.

Start a project
  • 17projects on GitHub, open to inspect
  • 5public-sector systems built and live online
  • Oct 2026incorporated in Kenya

Three ways to work with us

Use our infrastructure, pick a ready-made solution, or build something new with us from the first workshop to the published results.

AI as a Service

Deploy powerful AI capabilities instantly via our secure cloud infrastructure, removing the need for internal hardware management.

  • Compute that scales. Models run on managed cloud services that grow with demand and shrink when idle, so you pay for use, not for servers.
  • Managed data pipelines. We ingest, clean and schedule your data flows, from spreadsheets and databases to satellite imagery.
  • Model monitoring. Accuracy, drift, latency and cost are tracked, with alerts and a monthly report in plain language.
  • Your data, your region. Hosted on Google Cloud in regions you approve, or hybrid and on-premise where data must stay on site.
Discuss AI as a Service
Your apps, staff and citizensweb, mobile, USSD, WhatsApp
TovuTech API gatewaykeys, quotas, audit log
Models and agentsGemini, open models, your custom models
Data pipelinesingest, clean, Earth Engine, BigQuery
Secure cloud infrastructureencrypted, monitored, backed up
You use the top layer. We run everything beneath it.

AI Marketplace

Ready-to-use AI solutions built for African operating realities. Choose models, agents and workflows you can run on sovereign or hybrid infrastructure without starting from scratch.

  • Flood and drought early-warning agent
  • Trilingual citizen assistant
  • HR and performance workflow
  • Soil spectra analysis
  • Crop leaf disease check
  • Project monitoring map

Each one is adapted from a system we have already built. The marketplace opens in stages; ask about early access.

AI and GeoAI consultancy

Strategy-to-production delivery for teams that need real outcomes: use-case selection, data readiness, deployment, governance and performance.

  • AI readiness assessments and implementation roadmaps
  • AI risk management, ethics and data sovereignty
  • Staff training and capacity building
  • Compute and cost optimisation

We work in QGIS, ArcGIS Pro and Enterprise, Google Earth Engine, Python (PyTorch, scikit-learn, LightGBM, XGBoost, FastAPI), Vertex AI, BigQuery, Looker Studio, AppSheet and Firebase.

Which service fits?
You want toAI as a ServiceMarketplaceConsultancy
Use AI without buying hardwareYesYesOptional
Start in days, not monthsYesYesAfter discovery
Solve a problem nobody has packagedHosts itNoYes
Keep data on your own premisesHybrid optionHybrid optionYes
Get a published notebook or reportMonthly reportUsage reportYes
Train your own teamOnboardingOnboardingFull programme

Where we work

Choose a sector to see the problems we take on and work we have already done there.

How a project runs

Four stages, the same on every project. You see working output at the end of each one, so you can change direction early.

Projects and open science

Real systems for real problems, all on GitHub. Many were built by our founder for the County Government of Garissa before TovuTech existed; each card says where a project came from and how far along it is.

Follow new work at github.com/jmsmuigai. The TovuTech GitHub organisation goes live as projects move across.

We publish what the model can't do, too

Our soil nutrient research predicts 13 soil properties from mid-infrared spectra across four East African countries. The model was tested on soil it had never seen at three levels of difficulty. Performance falls as the test gets harder, and we report that openly, because a client deciding where to deploy needs to know.

  • 1,310 georeferenced soil samples from 16 sentinel sites
  • 1,794 model fits under group-disjoint cross-validation
  • 109 numbers in the paper, each re-derived by a verification script
  • 0.164 to 0.562 R² for Olsen phosphorus when calibration widened from three Kenyan sites to four countries
Read the code and paper Open the notebook
Best mean R² across 13 soil properties, by how much the test data was withheld

Higher is better; 1.0 would be perfect. Source: Mburu, Ogada and Kimwele, JAGST manuscript 1283, reproducibility package on GitHub.

Is your organisation ready for AI?

Answer five questions to get a starting point. Nothing you choose here leaves your browser.

Your result appears here as you answer.

The team

A small founding team, with specialist experts brought into each project as the work requires.

James Mukoma Mburu

Founder and Chief Executive Officer

James is an applied AI specialist, machine learning researcher and GeoAI practitioner with years of public-sector technology leadership. He is Director of ICT and E-Government at the County Government of Garissa, coordinates the county's Kenya Devolution Support Programme II (KDSP II), and chairs the County ICT Directors Caucus.

At the Department of Computing, Jomo Kenyatta University of Agriculture and Technology (JKUAT), his research covers hybrid CNN–RBF neural networks and multi-sensor fusion for soil nutrient prediction. He is PRINCE2 and ITIL certified, an alumnus of the Senior Leadership Development Programme at the Kenya School of Government and of Google's AI Agents Intensive, and has represented Kenya at ITE Singapore and the India AI Impact Summit in New Delhi.

He founded TovuTech in October 2026 to connect AI, machine learning and geospatial intelligence to practical problems in Africa.

  • Director of ICT and E-Government, Garissa
  • Chair, County ICT Directors Caucus
  • KDSP II coordinator
  • JKUAT researcher
  • PRINCE2
  • ITIL

GitHub

Kelvin Njiriri Mburu

Co-founder and Director

Kelvin is a co-founder and director of TovuTech Limited.

Responsible by design

The commitments we make on every engagement.

Data protection
We work to Kenya's Data Protection Act, 2019. Personal data is collected only when needed, encrypted in transit and at rest, and deleted when the purpose ends.
Data sovereignty
You decide where data lives. We write it down before we build, and offer hybrid and on-premise options for sensitive records.
Secrets stay secret
API keys and credentials live in encrypted server-side stores, never in web pages, notebooks or code repositories.
Honest models
We test on data the model has not seen, report where it fails, and keep a person in the loop for decisions that affect people.
Local context
Interfaces in English, Kiswahili, Af-Soomaali and other Kenyan languages, designed for low bandwidth and shared devices.
You own the result
Code, models, notebooks and documentation are handed over, so you are never locked in to us.

Questions people ask

What does TovuTech do?

We design and deliver AI solutions end to end: we define the workflow with you, design the data and machine learning pipeline, build and deploy the solution, and publish the analysis as a notebook or web page.

You're a new company. Why should we trust you?

TovuTech Limited was incorporated in October 2026. We don't claim a long client list. What we can show is the work: systems in daily use, research with every number re-checked by code, and repositories you can inspect before you talk to us.

Do we need our own servers to use AI?

No. With AI as a Service we host and run the models, pipelines and monitoring for you. If your data has to stay on site, we can deploy on your hardware or in a hybrid setup.

Where is our data stored?

Where you decide. Cloud hosting is in Google Cloud regions you approve; for records that must stay in the country or on your premises, we deploy there instead.

Can you publish our analysis?

Yes. Engagements can end with a reproducible notebook, an interactive web page, or both, published privately or openly on GitHub.

How long does a first project take?

A discovery workshop takes a few days. A first working pilot typically takes six to ten weeks, depending on how ready the data is.

Tell us about your idea

Describe the problem in a few sentences. We reply within two working days with questions, a suggested first step, or both.

Prefer to think it through first? can help you shape the brief.

This opens your email app with the message filled in. We use your details only to reply.