Completing Dec 2026 · Open to opportunities Internships · Graduate roles · Research · Full-time

SamuelMwania

Software EngineerMachine LearningData SystemsProduction Systems

I build machine learning and data systems that run in production: model, API, container, deployment, load test, retraining path.

BSc Software Engineering (ML) · African Leadership University · Completing December 2026
Based in Kenya · Internships and project work now · Full-time from January 2027

View work Resume Contact

Systems built end to end

ML services live in production

17th

Of 1,651 on Zindi

665

Automated tests on the capstone

Flagship · Capstone 2026

AegisVeritas

Two bidders. One owner.

Finds bidders in Kenyan public procurement that look independent but are linked to each other or to a debarred party, before the contract is signed. Built end to end, evaluated with pre-registered tests on Kenyan and Estonian data, and every flag shows its evidence.

0.146average precision in Kenya, against 0.067 for the baseline

0.85AUC on Estonian firms the model never saw

25/100top flags are real links; random finds about 2

Read the case study Watch the walkthrough

AegisVeritas evidence card for one supplier pair: likelihood of common control, the evidence behind it, what to check next and the verdict form. Real records, firm names masked.

Selected work

2024 – 2026

Research

Capstone and studies

Capstone · 2026 · Results in

Hidden supplier links in Kenyan public procurement

Pre-registered evaluation on Kenyan and Estonian data: average precision 0.146 against 0.067 for the baseline, and AUC 0.85 against 0.57 on unseen firms. Expert evaluation under way, under ALU ethics approval SEPT26-BEL-070.

Method and results →

Study · 2026

How much do hyperparameters decide DQN performance?

A 30-experiment controlled study with a public log. In my sweep, a discount factor of 0.90 beat the conventional 0.99.

Details →

Interests

Record linkage and graph ML
Risk analytics
AI for public institutions and finance
Low-resource African language NLP
Evaluation methodology

About

Software engineer in Kenya, building machine learning and data systems where software meets public institutions, finance and development.

I take a problem from raw data to a running service: evaluated against a stated baseline, deployed, tested, and documented so another engineer can extend it. Failed results are reported next to the ones that passed.

Every model has a baselinePre-registered testsDeployed and load testedLimitations reported

Education

BSc Software Engineering, Machine Learning

African Leadership University
Jan 2024 – Dec 2026 · graduation June 2027

Experience

Data Engineer Intern, Hello Tractor

Agritech platform linking smallholder farmers with tractor owners · Nairobi · Apr – Aug 2025. Prepared and audited datasets for reporting, and supported batch prediction runs.

Highlights

AegisVeritas capstone, supervised by Kevin Sebineza
17th of 1,651 on Zindi, five African languages
Two ML services live in production
Team work: Freeway DQN (3), MoMo Analyst (4, data pipeline)

Skills

PythonPyTorchTensorFlowscikit-learnLightGBMHugging FaceFastAPIPostgreSQLNeo4jDockerCUDANext.jsTypeScriptpytestPlaywrightLinux

Let's talk

Open to internships and project work now, and full-time roles from January 2027. Remote, or relocating where employer-sponsored opportunities are available. I reply within a day.

Copied to clipboard

Download resume LinkedIn GitHub

Contract or consulting work? Scope and rates