Tier 2 Plus · Elite Diploma
Artificial Intelligence and Machine Learning
A three-year elite diploma engineered to produce AI engineers who build the systems behind the APIs — not just the prompts in front of them.
- Code
- AIE 300
- Tier
- Elite Diploma (Tier 2 Plus)
- Duration
- 36 months · 144 weeks
- Cohort 01
- 02 November 2026
- Delivery
- On-Campus · Waterloo
- Credential
- WIATech Elite Diploma in Artificial Intelligence and Machine Learning
In 2026, anyone can call an LLM API. The shortage is of engineers who understand what they ship — who can debug a training run that diverges at step 4,000, explain why their RAG pipeline degrades on long queries, and defend their choice of fine-tuning method under research scrutiny. Our graduates build neural networks from scratch in NumPy before they touch PyTorch, and a miniature GPT end-to-end before they open Hugging Face. Engineers who only know the library cannot debug the library.
What you'll become
- AI Engineer
- Machine Learning Engineer
- Applied Research Engineer
- Research Engineer
- LLM Engineer
- MLOps Engineer
- AI Agent Engineer
- RAG & Search Engineer
- Computer Vision Engineer
- NLP Engineer
- Founding AI Engineer
- AI Research Resident
§ 01 · The difference
What sets this programme apart.
No Vibe Coding
Every architecture choice, every training run, every evaluation metric a graduate ships, they can defend. Copy-paste notebooks produce operators who cannot debug when the model diverges. We produce engineers who can.
Built from Scratch, Before Built from Library
Neural networks in NumPy before PyTorch. Backpropagation derived by hand before autograd. A miniature GPT end-to-end before Hugging Face. Engineers who only know the library cannot debug the library. By Semester 3, ours can.
Twenty-Four Oral Defences
One every six weeks. Each requires the engineer to defend their own architectural choices, training decisions, and evaluation results in front of faculty and external research evaluators, without AI assistance.
Research-Engineering, Not Tutorial-Following
From Semester 5, every engineer reads two AI research papers per week with cohort-submitted summaries. Three frontier papers are reproduced from PDF to working code and a baseline measurably improved — the discipline frontier AI labs hire for.
West African Problems, Frontier Standards
Engineering exercises target real African problems — Krio-language NLP, mobile-money fraud, agricultural ML, public-health surveillance, under-resourced-language transcription. Graduates engineer for their market, not translate Silicon Valley examples to it.
A Public AI-Engineering Identity
Every engineer ships under their own name — production systems, merged open-source contributions, technical writing on training runs and reproductions. Their GitHub is a working research-and-engineering record.
§ 02 · Who this is for
Built for engineers ready to operate at depth.
◆ The Working Engineer Specialising into AI
You are already a working developer or recent CS graduate. You want to specialise into AI engineering at depth — not collect another certificate, but operate at the level of senior AI engineers and applied researchers.
◆ The Mathematics-or-Physics Graduate
Your degree gave you the mathematical foundation but did not prepare you to ship production AI. You want to convert that foundation into engineering capability — and emerge with a portfolio of real systems, not just classroom proofs.
◆ The Self-Taught AI Builder
You have been training models on Kaggle, fine-tuning on weekends, reading Karpathy. You have hit the ceiling self-teaching always hits. You want the structure, the peer calibration, and the credential to operate at frontier-lab standards.
◆ The West African AI Pioneer
You want to build the AI infrastructure of this region — not emigrate to maintain someone else's. Multilingual NLP, agricultural ML, public-health AI, financial-services AI — the frontier work the region needs and the world will pay for.
§ 03 · The Tier 1 floor
What you need before you start.
This is an advanced programme that begins above the foundation floor. Clear it by passing the WIATech technical assessment — or by completing the matching Tier 1 Foundations courses. The assessment is a placement instrument, never a rejection.
Required foundations
Disciplined collaboration and a public engineering identity from day one. Every artefact ships via Git.
Memory, processes, and the machine — the substrate every serious AI lab assumes you already understand.
The algorithmic substrate of modern AI — including the structures underlying retrieval: vector search, nearest-neighbour, graph algorithms.
Most AI engineering work is data engineering. SQL fluency is assumed; the diploma builds production pipelines on it.
Linux is the AI engineer's operating environment — every training run, every GPU job, every deployment assumes it.
Python is the language of machine learning. Fluency is assumed at entry; the diploma operates Python as an engineering discipline.
Universal parallel co-requisite
Runs alongside the diploma from day one — no gate, no opt-out. It underwrites the RFCs, paper summaries, research writeups, and the twenty-four oral defences.
The one surviving bridge in the architecture, required only for AIE 300 (15 weeks). Cleared by assessment or completed before the diploma; it can run in parallel with the early diploma where scheduling allows.
AIE 300 gates on demonstrated capability and mathematical maturity, not paperwork. Applicants clear the Tier 1 floor (six foundations) plus the MTH 85 bridge — by assessment, or by completing what they are missing. The real prerequisite is the willingness to sustain frontier-depth work for three years. A student is shown only the foundations they are missing, never the whole catalogue.
Direct entry
Demonstrate the six foundations and the mathematical maturity on the aptitude assessment, clear the engineering interview, and enter AIE 300 in the November cohort.
Routed entry
Sit the assessment; for any gap, you are routed to the specific Tier 1 course (or the MTH 85 bridge) you need. Complete it, then proceed to the engineering interview and Tier 2 entry.
§ 04 · The architecture
6 semesters. One graduating engineer.
Year 01 · Foundations · Semester 01
Engineering & Computational Foundations
The professional engineering substrate that everything rests on — Linux and Python at engineering depth, the mathematics of ML taught computationally, SQL and data engineering, and the algorithms underlying modern AI: vector search, nearest-neighbour, graph algorithms.
Engineering Mindset & Computer Systems
Linux as the AI engineer's operating environment, Git internals and disciplined collaboration, and the professional practices that separate engineers from notebook-users.
Linux · Bash · Git · GitHub · Documentation Standards
Professional Python Engineering
Python at the depth Fluent Python teaches it — async, concurrency, packaging, performance, type systems, testing. Python as an engineering discipline, not a notebook scripting language.
Python · Async · Type Hints · pytest · Packaging · Performance
Data Structures & Algorithms for AI
DSA with explicit attention to the structures underlying modern AI — arrays, trees, hash tables, vector search, nearest-neighbour, and graph algorithms for knowledge graphs.
DSA · Vector Search · k-NN · Graph Algorithms · Big-O
The Mathematics of Machine Learning
Linear algebra, calculus, and probability taught computationally — vectors, matrices, eigenvalues, SVD, gradients, optimisation, distributions. The mathematics that determines whether a training run converges.
Linear Algebra · Calculus · Probability · Statistics · Optimisation
SQL & Data Engineering Foundations
SQL deeply — joins, window functions, query optimisation, relational design — and the data engineering pipeline from ingestion to feature production. The reality that most AI work is data engineering.
PostgreSQL · Window Functions · Query Optimisation · Data Pipelines
Software Engineering for AI Teams
Containerisation with the Tabempa 10-Step Dockerfile pattern, CI/CD, testing strategy for AI codebases, RFC and ADR writing, and code-review discipline — the shipping practices AI teams actually use.
Docker · GitHub Actions · pytest · RFCs / ADRs · Code Review
Capstone
The Production Data & Search System
Students build a production-grade data pipeline plus a working vector-search system over a real Sierra Leone or West African corpus — Dockerised, tested, version-controlled, documented, and defended orally.
Deliverables: Production data pipeline · Vector search system · Dockerised deployment · CI/CD pipeline with test gates · Engineering documentation · Oral engineering defence
Year 01 · Data & Classical Machine Learning · Semester 02
Data Engineering & Classical Machine Learning
NumPy from the internals, Pandas and Polars at scale, classical ML built from scratch — regression, trees, boosting, clustering — ML pipelines with MLflow/DVC/BentoML, and the first neural network built from scratch in NumPy with backpropagation derived by hand.
Scientific Python & NumPy Internals
NumPy from the internals — strides, broadcasting, memory layout, vectorisation — plus SciPy, Matplotlib, and Plotly. The performance engineering that decides whether a training run takes ten minutes or ten hours.
NumPy · SciPy · Matplotlib · Plotly · Vectorisation
Pandas & Polars at Production Scale
Pandas for analysis-grade work, Polars for production-scale tabular processing, data cleaning and validation, and EDA as a disciplined practice — the skills that fill 80% of working AI-engineering time.
Pandas · Polars · Data Validation · EDA · Pydantic
Classical Machine Learning — Built From Scratch
Linear and logistic regression from scratch, decision trees, random forests, gradient boosting, clustering, dimensionality reduction, and probabilistic methods — the classical foundation frontier engineers still reach for.
Regression · Trees & Boosting · Clustering · PCA / UMAP
Machine Learning Pipelines & Experiment Tracking
MLflow for experiment tracking, Weights & Biases for run management, DVC for dataset versioning, BentoML and FastAPI for serving, and orchestration with Prefect and Airflow — the infrastructure that makes ML reproducible.
MLflow · W&B · DVC · BentoML · FastAPI · Prefect
Statistical & Probabilistic Methods
Bayesian inference, hypothesis testing, A/B testing for ML systems, probabilistic graphical models, and Monte Carlo methods — the mathematics that determines whether a metric improvement is real or noise.
Bayesian Inference · A/B Testing · PGMs · Monte Carlo
Neural Networks From Scratch in NumPy
The pivotal module of the foundation year — neural networks built end-to-end in pure NumPy: forward pass, loss functions, backpropagation derived by hand, gradient descent variants, regularisation, batch normalisation. PyTorch is not used here.
NumPy · Backprop · Gradient Descent · Regularisation
Capstone
The Production Classical-ML System
Students ship a production classical-ML system solving a real Sierra Leone problem — mobile-money fraud detection, agricultural yield prediction, or WASSCE outcome modelling — with a full pipeline, experiment tracking, model serving, and monitoring, defended alongside the from-scratch neural network.
Deliverables: Production classical-ML system · Full MLflow + DVC pipeline · BentoML / FastAPI deployment · Monitoring & drift detection · Neural-network-from-scratch artefact · Oral engineering defence
Year 02 · Deep Learning · Semester 03
Deep Learning & Modern AI
PyTorch from the internals, GPU computing and distributed training, computer vision from CNNs to vision transformers, NLP from tokenisers to attention, the transformer built from scratch, and a miniature GPT built end-to-end.
PyTorch from the Internals
PyTorch as a professional tool — tensors, autograd, custom datasets and training loops, hooks and instrumentation, profiling and debugging. The depth to read the PyTorch source and modify it.
PyTorch · Autograd · Custom Modules · Profiling · Lightning
GPU Computing & Distributed Training
CUDA fundamentals, mixed precision training, memory management, Distributed Data Parallel, Fully Sharded Data Parallel, and tensor parallelism — the infrastructure literacy frontier AI labs assume.
CUDA · Mixed Precision · DDP · FSDP · GPU Profiling
Computer Vision Engineering
CNNs from scratch, transfer learning and modern fine-tuning, object detection with YOLO and DETR, segmentation, vision transformers (ViT, Swin, DINO), and production vision systems for African contexts.
CNNs · YOLO · DETR · Segmentation · ViT · DINO
NLP & Sequence Modelling
Tokenisation (BPE, WordPiece, SentencePiece), embeddings, RNNs and LSTMs for the historical foundation, and attention mechanisms built from scratch — understanding why the transformer replaced everything.
Tokenisation · Embeddings · RNN / LSTM · Attention
The Transformer — Built From Scratch
The pivotal module — the transformer implemented end-to-end from scratch: multi-head attention, positional encoding, layer normalisation, residual connections, feed-forward layers. Engineers who only know the Hugging Face API cannot debug it; ours can.
Transformer · Multi-Head Attention · Positional Encoding · Layer Norm
Modern Decoder Models & Inference Engineering
Modern decoder-only models, scaling laws, production tokenisers, KV-cache optimisation, quantisation (INT8/INT4/GPTQ/AWQ), and inference engines including vLLM — culminating in a miniature GPT built end-to-end.
Mini-GPT · KV-Cache · Quantisation · vLLM · Scaling Laws
Capstone
The Deep Learning System in Production
Students ship a trained or fine-tuned deep learning model in production — CV on Sierra Leone data, an NLP system on an under-resourced African language, or a miniature GPT trained end-to-end — with a full training pipeline, distributed training where required, and quantised inference, defended orally.
Deliverables: Trained deep learning model · Full training infrastructure · Quantised production inference · Experiment tracking record · Distributed training artefact · Oral engineering defence
Year 02 · Production AI · Semester 04
Production AI Engineering
LLM application engineering with structured outputs and tool use, production-grade RAG from naive baseline through hybrid retrieval with reranking, AI agents with planning and memory, MLOps at scale, Kubernetes for ML, AI security, and cost engineering.
LLM Application Engineering
Prompt engineering as a disciplined practice, structured outputs with Pydantic and JSON schema, tool use and function calling, embeddings and vector databases at scale, cost and latency engineering, and eval-driven development that gates deployment.
Anthropic Claude · Structured Outputs · Tool Use · Embeddings · Evals
Production RAG Engineering
RAG from naive baseline through production-grade hybrid retrieval — chunking strategies, embedding selection and fine-tuning, vector databases (pgvector, Qdrant, Weaviate), reranking with cross-encoders, and retrieval evaluation.
RAG · pgvector · Qdrant · Hybrid Retrieval · Reranking · Retrieval Evals
AI Agent Engineering
Agents as a discipline — ReAct and planning patterns, memory systems (short, long, hierarchical), multi-agent orchestration, trajectory evaluation, and agentic coding assistants.
ReAct · Planning · Memory Systems · Multi-Agent · Trajectory Eval
MLOps & ML Infrastructure
CI/CD for ML with model validation gates and progressive deployment, drift detection, model monitoring, feature stores, real-time serving, and Kubernetes for ML with GPU scheduling and autoscaling.
Kubernetes · Feature Stores · Drift Detection · Distributed Training
AI Security & Responsible AI
Prompt injection defence, data poisoning, model theft, threat modelling for AI systems, and responsible AI in practice — bias evaluation, content filtering, jailbreaking resistance, and AI-specific incident response.
Prompt Injection · Data Poisoning · Threat Modelling · Bias Eval
AI System Design & Cost Engineering
AI system design at scale — multi-region, latency-engineered, cost-optimised — plus cost engineering as a senior discipline: tokens per request, batch efficiency, caching, and model-selection economics.
System Design · Cost Engineering · Latency · Capacity Planning
Capstone
The Production AI Product
Students ship a production AI product solving a real West African problem — a Krio-language assistant, a mobile-money fraud-detection system, a healthcare triage assistant — with production RAG or agent architecture, full MLOps, a security audit, and cost engineering, defended before faculty and external AI evaluators.
Deliverables: Production AI product · Full eval harness with gates · MLOps infrastructure · Security & responsible-AI documentation · Cost engineering report · Oral engineering defence
Year 03 · Frontier & Research · Semester 05
Frontier AI & Research Engineering
Reading AI papers as engineering specifications, reproducing frontier papers from PDF to code, advanced architectures (Mixture of Experts, state-space models, long-context attention), generative AI (diffusion, multimodal, audio, video), and reinforcement learning with RLHF.
Research Engineering & Paper Mastery
From Week 97, every engineer reads at least two AI papers per week with summaries and critical reactions submitted to the cohort — paper reading as engineering specification, and identifying contribution versus noise.
Paper Reading · Research Methodology · Reproducibility · Ablation Design
Reproducing Frontier Papers
The pivotal module of the research year — three frontier papers reproduced from PDF to working code with no GitHub repos consulted until results match, plus ablation studies and experiment design.
Reproduction · Ablation · Experiment Design · Reproducibility
Advanced Architectures
Mixture of Experts, sparse models, state-space models including Mamba, long-context attention (Ring Attention, Flash Attention), and self-supervised and contrastive learning — the post-transformer frontier.
MoE · State-Space Models · Mamba · Flash Attention · Contrastive Learning
Generative AI — Diffusion & Multimodal
Diffusion models through modern latent diffusion, GANs in context, CLIP and vision-language systems, audio with Whisper and modern TTS, and video generation foundations — the generative frontier across modalities.
Diffusion · Latent Diffusion · CLIP · Whisper · Audio & Video
Reinforcement Learning & RLHF
Markov Decision Processes, deep reinforcement learning, RLHF and preference learning for language models — the technique behind modern LLM alignment — and a focused survey of robotics and embodied AI.
MDPs · Deep RL · RLHF · DPO · Preference Learning
Frontier AI Operations
Operating frontier AI systems in production — hallucination measurement and mitigation, long-running evaluation pipelines, multi-model serving, and capacity engineering for frontier workloads.
Hallucination Eval · Multi-Model Serving · Capacity Engineering
Capstone
The Frontier Paper Reproduction
Each engineer reproduces a frontier AI research paper from PDF to working code — no GitHub consulted, no shortcuts — then improves the baseline measurably, defended before faculty and external research evaluators. The proof a graduate can read the AI literature like an applied researcher and ship from it.
Deliverables: Reproduced frontier paper with matching results · Reproducibility documentation · Measurable baseline improvement · Ablation study package · Public technical writeup · Research-engineering defence
Year 03 · Residency · Semester 06
Senior AI Residency & Final Capstone
Engineering leadership for AI teams, AI-specific incident response, a production AI residency with on-call drills for ML systems, and the twelve-week dual-track final capstone — an open-source contribution to a major AI project OR a publishable research project — defended in 60 minutes before an external panel.
Engineering Leadership for AI Teams
RFCs and ADRs for AI systems, design documents, and technical writing as a senior engineer's defining skill — the communication discipline that decides whether an engineer's work compounds across a team.
RFCs · ADRs · Design Docs · Technical Writing
Code Review & Estimation for AI
Code-review mastery for ML codebases — catching silent statistical bugs and evaluation flaws — plus estimation and planning for AI projects and the reasons ML estimates are systematically wrong, with mentorship and teaching.
ML Code Review · Estimation · Mentorship · Teaching
Production AI Residency I — Incidents
Real team rituals — daily standups, on-call drills for ML systems with injected failure modes (hallucination spikes, prediction drift, training instability, cost explosions), postmortems, and Incident Commander rotation.
On-Call · Incident Response · Postmortems · AI Failure Modes
Production AI Residency II — Performance & Cost
Performance and cost engineering at production scale — reading the bill of a real AI workload, optimising serving costs, AI engineering ethics, and the sustainability of frontier-scale workloads.
Performance · Cost Engineering · AI Ethics · Sustainability
Final Capstone — Selection & RFC
The first half of the final capstone. Track A — a substantive contribution to a major AI open-source project (PyTorch, Hugging Face, vLLM, LangChain, OpenSearch). Track B — a research project with publishable results. The RFC is written and maintainers or advisors engaged.
Open-Source Selection · Research Proposal · RFC · Maintainer Engagement
Final Capstone — Implementation & Defence
Implementation across multiple sprints — for Track A, code-review iteration with maintainers and a merged PR; for Track B, completed experiments, ablation studies, and a written paper — plus a published blog post, career launch, and a 60-minute panel defence.
Implementation · Defence · Publication · Career Launch
Capstone
Open-Source AI Contribution — or — Publishable Research
The graduation requirement that defines a WIATech AI engineer. Track A: a substantive merged contribution to a major AI open-source project. Track B: a research project with publishable results and a fully reproducible experiment package. Either is defended in 60 minutes before an external industry and research panel.
Deliverables: Track A: Merged PR to a major AI open-source project · Track B: Reproducible research paper with results · Public technical writeup under the engineer's name · Polished public AI-engineering identity · 60-minute industry & research panel defence
§ 05 · The toolkit
The stack you'll master.
§ 06 · Grading
How the work is measured.
§ 07 · Credentials & career
What you walk out with.
A Tier 2 Plus Elite Diploma — AIE 300 — in Artificial Intelligence and Machine Learning, issued by the Waterloo Institute of Advanced Technology — an academy of Tabempa Engineering Limited. It is accompanied by the verified portfolio of 144 shipped deliverables, the three reproduced frontier papers, and the defended dual-track capstone that defines a WIATech AI engineer.
The portfolio
- 144 weekly deliverables shipped publicly on GitHub
- Six production-grade capstone systems
- Three reproduced frontier AI papers with public writeups
- A merged AI open-source contribution OR research artefact
- Engineering log spanning the full 144 weeks
- Twenty-four oral defence records on file
- West African AI-engineering impact record
Career acceleration
- Mock interviews — system design, ML breadth, research
- Senior-track resume & LinkedIn AI-engineering brand
- GitHub profile polishing & research portfolio
- Remote work, freelancing & international engagement
- Direct introduction to partner AI organisations
- Alumni network access — for life
§ 08 · Admissions
Who we admit. How we admit them.
The most selective intake the institute runs. We do not admit on credentials — no WASSCE results, university degree, or formal AI certifications are required. What we require is evidence of substance and the mathematical maturity to take on frontier work: a GitHub profile with real code, a model you trained, a paper you read deeply, a project that runs.
Application
Online application form, GitHub profile (or equivalent portfolio), and a 300-word statement of intent answering why senior-track AI engineering rather than a faster credential.
Aptitude Assessment
A WIATech-administered test in mathematical reasoning, programming literacy, problem decomposition, and written communication. No prior ML knowledge required.
Engineering Interview
A structured interview with faculty evaluating mathematical maturity, engineering judgement, intent, and the capacity to commit to 144 weeks of substantive work — not existing credentials.
Offer & Enrolment
Successful applicants receive a formal offer, enrolment package, and onboarding schedule.
Starting from the foundations, via the AI & Machine Learning Foundations Pathway: NLe 157,500 total (NLe 22,500 foundations + NLe 135,000 diploma). Tier 2 Plus Elite Diploma tuition is set in advance and paid in monthly instalments after a seat deposit. Full tuition & payment →