● AI/ML Engineer . Data scientist
Hi, I'm Thamu. As an AI/ML engineer and data professional, I have had to wear many different hats, but ultimately my goal is to add value in any way I can. Views here and in this blog are my own.
This site is where I write about applied AI, machine learning engineering, and data engineering through the lens of building end-to-end systems: what works, what breaks, and what I learn from taking ideas into implementation.
I also help organise my local Google Developer Group (GDG), which keeps me close to builders, students, engineers, and community conversations around emerging technology. Outside of work, I am learning German and preparing to write the A1 exam. The goal is to share useful ideas, collaborate with people working on meaningful problems, and keep learning in public.
// 02 — Writing
A practical guide to MLflow autologging: one-line experiment tracking, framework-specific integrations, GridSearchCV child runs, and production guardrails.
How to make MLflow experiment tracking reproducible by pairing runs, metrics, model artefacts, signatures, and registry versions with a UV lockfile.
How to evaluate RAG and agentic GenAI systems with MLflow 3.x, mlflow.genai.evaluate(), LLM-as-a-judge scorers, custom trace-aware metrics, and prompt A/B tests.
Ein praktischer Leitfaden zu MLflow Autologging: One-Line Experiment Tracking, Framework-Integrationen, GridSearchCV Child Runs und Produktions-Leitplanken.
Wie man MLflow Experiment Tracking reproduzierbar macht, indem Runs, Metriken, Modellartefakte, Signaturen und Registry-Versionen mit einem UV Lockfile verbunden werden.
Wie man RAG- und agentische GenAI-Systeme mit MLflow 3.x, mlflow.genai.evaluate(), LLM-as-a-Judge-Scorern, eigenen trace-bewussten Metriken und Prompt-A/B-Tests bewertet.
Un guide pratique de MLflow autologging : experiment tracking en une ligne, integrations framework, child runs GridSearchCV et garde-fous de production.
Comment rendre le suivi d'experiences MLflow reproductible en combinant runs, metriques, artefacts, signatures, versions de registre et lockfile UV.
Comment evaluer les systemes RAG et agentiques avec MLflow 3.x, mlflow.genai.evaluate(), des scorers LLM-as-a-judge, des metriques trace-aware et des tests A/B de prompts.
// 03 — What I do
Sitting at the intersection of business and AI, I try to focus on the solution first. The aim is to reach for the right tool for the job, whether that's generative AI, classical machine learning, or a straightforward heuristic. Most of the projects I'm part of run end-to-end, from helping frame the problem with stakeholders, designing and implementing the infrastructure, to monitoring, maintaining, and supporting new features once they're in production.
RAG systems and LLM agents where generative AI genuinely earns its place; classical ML, mathematical optimisation, or statistical methods where they do the real work underneath. In practice that has meant forecasting and demand modelling, operations optimisation, and retrieval-augmented assistants over internal data.
Training pipelines, model serving, drift monitoring, retraining, and governance work. I believe that the principles don't change with the stack, for example solid orchestration, automated testing, and observability that catches problems before users do.
Translate business outcomes into data and AI architectures. I sit between stakeholders and the model, framing the question before a single line of code is written.
Design reliable, cost-aware infrastructure for AI workloads, currently across Azure and Databricks and previously GCP. Terraform-certified, comfortable making infrastructure decisions.
// 04 — Background
// education
Thesis work in stochastic processes studying the stochastic area. Some of the results from this work contributed towards the paper below.
📄 Publication
Large deviations of the stochastic area for linear diffusions Johan du Buisson, Thamu D. P. Mnyulwa, and Hugo Touchette. Published in Physical Review E 108, 044136 (2023).
Courses: Data mining, Time series, Extreme value theory, Multivariate Analysis, Simulation theory, Biostatistics
Courses: Statistics, Mathematics, Investment management, Economics, Accounting, Finance
// 05 — Certifications
Google Cloud
Google Cloud
Google Cloud
HashiCorp
Microsoft
Microsoft
Microsoft
McKinsey.org
// 06 — Projects
// Transmit
Reach me directly about collaborations or any problem you are working on.
