Self-directed curriculum

Everything
you are learning

A personal library of courses built for deep, independent study. Each one is fully self-contained — read, practice, build. No internet required except where noted.

6
Courses
24
Weeks · Speech
9
Sessions · AI Programme
~1
Working LLM
AI & Product Strategy
AI · Product 9 sessions
Building Production-Ready AI Products for African Markets
The complete 9-week programme for founders and technical leaders. From product strategy through MLOps and AI economics — every session self-contained with worked fintech examples and an interactive deliverable.
Strategy Architecture MLOps UX Safety Economics
↓ 18 hr total
AI · Math
Math for AI Products
The six mathematical concepts behind the AI programme — probability, distributions, vectors, model evaluation, statistics, and cost modelling. Interactive throughout. Prerequisite for Sessions 2, 3, 4 and 8.
Probability Embeddings AUC Cost models
→ ~2 hr
AI · Agents New
AI Agents, from understanding to deployment
What an agent really is, the 2026 trends and the genuine concerns — reliability, security, cost — and how to build and deploy one safely. From the reason-act-observe loop and MCP to multi-agent patterns and production guardrails.
MCPMulti-agentGuardrailsEvaluation4 simulators
→6 modules
Engineering
Engineering · Python
Building Large Language Models from Scratch
Build a working Transformer language model in ~280 lines of PyTorch. The same architecture as GPT-4, at a learnable scale. By the end you train it on real text and generate output from it.
Transformers Attention PyTorch Backprop Generation
→ ~5 hr
Credit & Risk
Credit · Foundation Models Live demo
PRAGMA: Behavioural Foundation Models
Revolut trained one model on raw banking events and beat their task-specific pipelines. Distil what PRAGMA is, then build a scaled-down version for customer segmentation — with runnable PyTorch, a live clustering demo, and an honest take on what it takes for us to build one.
Foundation Models Embeddings Segmentation Masked Modelling PyTorch
→ 1 week
Credit · Foundation Models Live demo
PRAGMA: Behavioural Foundation Models with Production Deployment
Revolut trained one model on raw banking events and beat their task-specific pipelines. Distil what PRAGMA is, then build a scaled-down version for customer segmentation — with runnable PyTorch, a live clustering demo, and an honest take on what it takes for us to build one.
Foundation Models Embeddings Segmentation Masked Modelling PyTorch
→ 1 week
Credit · Foundation Models New
TREASURE: the Visa Payment Foundation Model
Visa's transaction foundation model, distilled — decoder-only and autoregressive, predicting both cardholder behaviour and payment-network signals. +111% fraud detection, +104% recommendations. Contrasted throughout with PRAGMA.
Decoder-only InfoNCE Network signals Embeddings vs PRAGMA
→ 1 week
Credit · Foundation Models New
TREASURE: Production Deployment
From paper to a live scorer on our stack — lakehouse data pipeline, ONNX export into the decisioning engine, millisecond real-time scoring, embedding serving, and a staged, monitored, governed rollout. Tailored to in-country residency and the DS/SWE ownership split.
ONNX serving Real-time Lakehouse Rollout Governance
→ companion
Math & AI
Credit · Math
Mathematics of Credit Scorecards
The complete math stack for scorecard development — statistics, Weight of Evidence, logistic regression, XGBoost, AUC, KS statistic, scorecard scaling with PDO, and production monitoring with PSI. 8 modules, 12 interactive demos.
WoE / IV Logistic Regression XGBoost AUC / KS PSI Scorecard Scaling
→ ~4 hr
AI · Math
Transformer Math
Eight rungs, from "what does ℝB×L×d mean" to reading a real survival-analysis loss. No prerequisite beyond arithmetic. By the end, the formulas in a modern foundation-model spec — dot products, softmax, cross-entropy, InfoNCE, hazards — read like sentences.
8 rungs, each depending on the last 5 interactive widgets — drag the math 0 prereq beyond multiply, add, and logarithms 1 goal read real research formulas unaided
→ ~X hr
Math
Foundational Math
Powers, logarithms, trigonometry, functions, the numerical-analysis toolkit — and now calculus, probability and statistics too. No fluff, no proofs for their own sake: intuition first, then a worked example you can follow line by line. The "oh right, that's what it means" course.
9 days, arithmetic through statistics 7 interactive widgets to poke at 3 on numerical analysis, calculus & probability 40+ worked examples with the arithmetic shown
→ ~X hr
Infrastructure
Infra
Operating systems, from the metal up
What actually happens between your code and the hardware. Four weeks on the three jobs every OS does — sharing the CPU, sharing memory, and storing things that survive a power cut — ending where it matters to you: how virtual machines and containers really work.
4 weeks: CPU, concurrency, memory, storage 4 interactive simulators to run 3 illusions the OS maintains for you ∞ bugs this will help you finally understand
→ ~X hr
Infra
Cloud infrastructure, from instance to Kubernetes
What you're actually renting when you launch a server, how the network and firewalls around it work, what drives the bill, and how containers and Kubernetes run on top — then how to stand up cloud infrastructure yourself, public or self-hosted. Hands-on, with the real commands.
6 modules, instances → your own cloud 4 interactive simulators 2 paths: public cloud & OpenStack ∞ dollars this will help you stop wasting
→ ~X hr
Infra
Linux, from foundations to a hardened production host
The operating system your products run on, understood properly. Where Linux came from and how it got here, how it's actually built, and the four skills that matter in production: managing it, tuning it, hardening it, and configuring it — with Ubuntu and RedHat covered side by side.
6 modules: foundations → hardened host 4 interactive tools to try 2 families, Ubuntu & RedHat, throughout 100% real commands you can run today
→ ~X hr
Voice AI
Voice AI New
Voice AI, from pipeline to orchestrator
How a machine holds a real-time spoken conversation — turns, latency, interruptions and code-switching, tuning for Swahili and other African languages with Chatterbox, and running it on your own hardware.
ChatterboxSwahiliTurn-takingOn-prem4 simulators
→6 modules
Voice AI
End-to-end speech models
The other architecture — speech-to-speech models. How they work (neural codecs, audio tokens), how they are tuned, and a step-by-step look at exactly what it would take to make one work for our languages and use cases.
Audio codecsThinker-TalkerFine-tuningRoadmap
→5 modules
Voice AI New
Local TTS on cheap hardware — Piper, Kokoro & XTTS
Three lightweight open-source voices that run offline on a Raspberry Pi or a cheap CPU box. Their architectures, how to tune them for Tanzanian Swahili, the licensing that decides which you can actually ship, and how to deploy for next to nothing.
Piper Kokoro XTTS Swahili On-prem
→6 modules
+
Add a new course
Drop any .html course file into this folder.
Then add a card below using the template in the comment at the bottom of this file.