Certificate programme
Machine Learning Engineer
Training a model that scores well in a notebook has become accessible to many people. Running that model every day, on real data that changes, behind an interface other applications can call, with monitoring that warns when quality degrades and a documented retraining procedure, remains the work of a handful of professionals.
- Duration
- 5 weeks, 10 sessions, 50 hours
- Format
- In person or online
- Credential
- Training certificate
Tuition195,000 XOF
By the end of the programme, you will be able to
- Frame a business need as a learning problem, with its target, its metric and a simple baseline
- Structure a Python project that is version-controlled, tested and runs in one command
- Prepare real data and build explanatory variables without leaking information
- Train, compare and tune regression, classification, ensemble and forecasting models
- Choose a decision threshold from the cost of the error, and track your experiments in MLflow
- Classify text and query a corpus with a language model, citing its sources
- Serve a model behind a FastAPI interface, containerise it with Docker and monitor it
- Explain a model's decision, document its biases and limits, and defend your project before a panel
Who it is for
Target audience
The certification is for people who can already analyse data and want to take the step into production: data scientists, advanced data analysts, developers, statisticians, engineers, graduates in computer science, statistics or applied mathematics.
Prerequisites
Four conditions are required:
- Hold at least a three-year degree (BAC+3) in statistics, computer science, mathematics, quantitative economics or engineering, or equivalent professional experience
- Be proficient in Python at the level of data structures, functions, pandas and file handling; the Waaldé Impact Data Scientist certificate or Python course meets this prerequisite
- Have a solid grounding in statistics: distribution, correlation, regression, the notion of a test
- Have a laptop with at least 8 GB of RAM, preferably 16 GB, and 40 GB of free disk space
A fifteen-minute technical interview, with a short coding exercise, is held before any enrolment. A candidate who does not yet program in Python will be directed to the Python course and then to the Data Scientist certificate, which genuinely prepare for this programme.
Teaching method
The certificate runs over five weeks, with two sessions a week, on Saturday and Sunday from 8.30 a.m. to 1.30 p.m.: ten five-hour sessions, 50 supervised hours in all. Personal work comes to two or three hours a week, and three to four during the project week.
Each session is preceded by video capsules to watch remotely, about an hour and a quarter: the theory and the demonstrations are watched beforehand, and the session itself is given over to supervised hands-on practice and exercises on real data. Each session also has a "going further" section covered in the course material.
The course is built the other way round from the usual ones: every model learned is immediately turned into a working service. Sessions can be followed in person in Ouagadougou or online and live, as each participant prefers, at the same price. The cohort is limited to twelve, with individual supervision on the project, and a fifteen-minute technical interview — including a short coding exercise — comes before any enrolment.
Every participant leaves with:
- The video capsules for every session and the complete course material in PDF, module by module
- A reference code repository holding every model and pipeline built in class
- A project template ready to clone: training, pipeline, FastAPI interface, Docker, monitoring
- Annotated notebooks on each family of models
- A cheat sheet for scikit-learn, boosting, PyTorch and MLflow
- A framing grid for a learning problem and a model audit grid
- A model card template and a pre-production checklist
- A guide to preparing for machine learning engineering interviews
The libraries taught are free and open source, and the exercises are sized to stay within free compute tiers.
Detailed program
10 sessions · 50 h of teaching
Week 1 — The profession, the code and the data
2 sessions · 10 h- Turn three real needs into learning problems: target, unit of observation, horizon
- Choose the metric according to the cost of the error, and compare it with a simple business rule
- Check the local constraints: real data volume, connectivity, cost of compute
- Exercise: write the problem sheet for three use cases and have it reviewed by a partner
- Going further, in the course material: reinforcement learning; the detailed framing grid; the local job market
- Turn a notebook into reusable modules and functions
- Create a virtual environment, pin the dependencies and separate configuration from code
- Work with a Git repository: commit, branch, pull request
- Write two automated tests on a data preparation function
- Exercise: make the project run in a single command, driven by a configuration file
- Going further, in the course material: detailed logging, type hints, style conventions and continuous integration
Week 2 — Data and supervised learning
2 sessions · 10 h- Audit a real dataset: volume, quality, granularity
- Handle missing values and categorical variables, then scale
- Split into training, validation and test sets, and spot an information leak
- Handle class imbalance through weighting and compare two approaches
- Exercise: turn a raw dataset into a documented, leak-free training matrix
- Going further, in the course material: geographic data, aggregated data and rolling windows, synthetic data
- Beat a baseline linear regression with Ridge or Lasso, then analyse the residuals
- Train a logistic regression and read its coefficients as odds ratios
- Compute precision, recall and F1, then choose a threshold from the business cost of the error
- Run a stratified cross-validation and tune with random search
- Exercise: build a classifier on imbalanced data and justify the threshold you keep
- Going further, in the course material: support vector machines, naive Bayes, probability calibration, nested validation
Week 3 — Ensembles and time series
2 sessions · 10 h- Train a random forest and read variable importance with due caution
- Tune a gradient boosting model with early stopping on the validation curve
- Build a reusable scikit-learn pipeline
- Log parameters, metrics and artefacts in MLflow, then compare the runs
- Exercise: optimise a boosting model on a concrete case and track every experiment
- Going further, in the course material: XGBoost, LightGBM and CatBoost compared; model stacking and weighted voting
- Decompose a monthly series: trend, seasonality, stationarity
- Forecast with exponential smoothing, then with Prophet, and compare the errors
- Turn the series into a supervised problem: lags, rolling windows, calendar variables
- Validate with a rolling origin, not with ordinary cross-validation
- Exercise: forecast a real series over twelve months and compare three approaches
- Going further, in the course material: ARIMA and SARIMA, prediction intervals, multi-horizon forecasting
Week 4 — Text and putting models into service
2 sessions · 10 h- Clean a corpus of messages and build a TF-IDF representation
- Classify messages with a classical baseline, then with a fine-tuned pre-trained model
- Call a language model through its API: parameters, costs, rate limits
- Chunk, index and retrieve passages to obtain a sourced answer
- Exercise: build a document assistant on an internal corpus and count its invented answers
- Going further, in the course material: named entity recognition, computer vision, agents and tool calling, open models running locally
- Serialise a model and pin the versions of the libraries it uses
- Build a prediction API with FastAPI: input validation, errors, documentation
- Containerise the service with Docker, then deploy it on a token-protected virtual machine
- Scope the project: chosen case, target, metric, baseline and performance threshold
- Exercise: deliver a containerised prediction API and submit the scoping sheet
- Going further, in the course material: batch serving, response-time optimisation, end-to-end tests
Week 5 — Building the project and defending it
2 sessions · 10 h- Set up monitoring for the service: drift indicators and alerts
- Explain a prediction with permutation importance or SHAP, then in plain language
- Measure the model's performance on two subgroups and record the trade-offs
- Exercise: finalise the model card, with a one-to-one session with the trainer
- Going further, in the course material: advanced statistical detection, legal responsibility when a model errs, the place of human decision
- Present the project in ten minutes: problem, data, model, performance, deployment, limits
- Demonstrate live the API call, a prediction and the monitoring
- Peer code review between participants, then defence of the choices before the panel
- Exercise: answer five minutes of questions, then receive the certificate at the end of the session
- Going further, in the course material: preparing for the technical interview, the local and remote job markets
Assessment and certification
The Machine Learning Engineer certificate is awarded on an admission average out of 20, made up of three components:
- The certification exam (50%) — 25 questions drawn at random from a bank of more than eighty, in 90 minutes. Two attempts are allowed, seven days apart, and the better one counts.
- The deployment project (30%) — defended in the tenth session.
- Attendance (20%) — recorded session by session. An absence announced at least three days in advance counts as half; an unannounced absence does not count.
The admission average is 12 out of 20. A mark below 8 out of 20 in the exam is eliminatory: neither the project nor attendance can make up for it. As soon as the average is reached, the certificate is issued automatically and becomes available in the participant's account.
The certificate is issued by Waaldé Impact, signed by the trainer, and states the course title, the 50 hours, the period, the list of modules completed and the subject of the deployment project presented. Each certificate carries a unique number and a verification code.
Participants who do not reach the average receive a certificate of attendance. They may retake the exam and present their deployment project with the following cohort, at no extra cost.
Participants also receive:
- The complete course material in PDF format, module by module
- A reference code repository containing all the models and pipelines built in class
- A ready-to-clone template project: training, pipeline, FastAPI interface, Docker, monitoring
- Annotated notebooks on each family of models
- A cheat sheet for scikit-learn, boosting, PyTorch and MLflow
- A learning problem framing grid and a model audit grid
- A model card template and a pre-production checklist
- A guide to preparing for machine learning engineer technical interviews
Upcoming sessions
Next dates being prepared
Leave your email: we will let you know as soon as registration opens.
Frequently asked questions
195,000 XOF per participant. Training for a team or an organisation is quoted on request.
5 weeks, 10 sessions, 50 hours.
Yes: you can attend onsite or online, as you prefer.
Four conditions are required: hold at least a three-year degree (BAC+3) in statistics, computer science, mathematics, quantitative economics or engineering, or equivalent professional experience ; be proficient in Python at the level of data structures, functions, pandas and file handling; the Waaldé Impact Data Scientist certificate or Python course meets this prerequisite ; have a solid grounding in statistics: distribution, correlation, regression, the notion of a test ; have a laptop with at least 8 GB of RAM, preferably 16 GB, and 40 GB of free disk space ; a fifteen-minute technical interview, with a short coding exercise, is held before any enrolment. A candidate who does not yet…
Yes: participants who meet the completion requirements receive a Waaldé Impact certificate with a unique number that employers can verify online.
Fill in the online enrolment request: no payment is taken on the site. The firm reviews your application, then contacts you to confirm your place and the payment terms. No session is scheduled yet: leave your email on this page to be told when registration opens.
Join the next cohort
Cohorts are deliberately small to guarantee genuine individual support. Reserve your place or ask us your questions.