A grid of 273 small squares standing for 2,730 learners, 124 of them lit in cyan for the 1,239 who finished the course, beside a twelve-hour dial

Twelve Hours and 1,239 Certificates: What Kenya's Civil-Service AI Course Delivered

The short version The Kenya School of Government has turned its e-learning arm into an institute for digital and emerging-technology training, the e-Learning and Information Technology Institute. T...

Oct 18, 2026 AI in Africa, AI Engineering

A query card for account 8890 being sorted into a cache slot that holds account 4471's answer, stamped WRONG

The Cache Was Confident and Wrong: Choosing a Semantic Cache Threshold You Can Defend

A semantic cache is the only component in your stack whose failures are recorded as savings. A hit is logged as an avoided call, a miss is logged as cost, and nothing in the default telemetry recor...

Oct 17, 2026 LLM, AI Engineering

A row of words in different languages all written in the same Arabic script converging on a classifier rack of many labelled slots, with one slot lit and a live waveform feeding a 480 millisecond tag

Same Script, Fifty-One Answers: What Morocco's Open Darija Release Actually Solves

In one paragraph: On Tuesday 29 September 2026, Morocco’s Ministry of Digital Transition and Administrative Reform released two open-source language tools built with Mistral: a 51-class Arabic lang...

Oct 16, 2026 AI in Africa, LLM

Three coloured pens stacked over a column of message cards, most cards carrying matching labels and two rows where one pen has drifted off to a dashed disagreement rail, beside a readout showing a kappa of 0.63 and an agreement ceiling of 0.74

Agreement Before Accuracy: Auditing the Labels You Are About to Train On

Two numbers from the fixture in this post: three annotators labelling 240 support messages match each other on 74 to 79 percent of the items they both labelled, and an oracle that copies annotator ...

Oct 15, 2026 Machine Learning, ML Ops

A dark retinal image at left with vessel branches, an optic disc and small red lesions, a purple tag reading SELENA+ v1.7.0 offline, and two diverging pathways to the right, a cyan one reaching 10.9 days and 35.9 percent attendance and a dashed yellow one looping out to 36.9 days and 18.2 percent

The Model Was Not the Intervention: Reading Tanzania's Retina Screening Trial

Introduction The trial, in one line On 13 October 2026 The Lancet Global Health published a randomised controlled trial of AI-supported diabetic retinopathy screening across 24 hospitals in Tan...

Oct 14, 2026 AI in Africa, Machine Learning

A wide stream of candidate vectors passing a mesh labelled WHERE, with only a few dots surviving below and a small cluster of the filtered rows sitting far off to the side

Your WHERE Clause Ate the Recall: Sizing the Candidate Pool Behind Filtered Vector Search

A vector index returns candidates. The filter decides how many of them answer. pgvector’s documentation gives the arithmetic: with the default hnsw.ef_search = 40 and a condition matching 10% of ro...

Oct 13, 2026 Machine Learning, ML Ops

Globe with glowing AI nodes across seven regions

Tuesday AI Update: Oct 13, 2026 — Meta's First Frontier Model Arrives Behind a Paywall

The week in AI October 6 to 13 was the week three of the four US frontier labs moved within 48 hours of each other. Meta shipped its first frontier model, Microsoft put its own inference silicon o...

Oct 12, 2026 AI Engineering, AI in Africa

A grid map of African member states with 44 tiles lit and three greyed out, a radio tower feeding a speech-to-text wave into a signal funnel, a scatter of duplicate signal dots collapsing into single event markers, and a detection clock whose median marker sits at day one while the tail runs past day seven

Africa's Outbreak Radar Reached 44 of 47 States: Reading What the Expansion Buys

Introduction The report, in one line The WHO African Region’s October 2026 health emergency intelligence and surveillance report says its Epidemic Intelligence from Open Sources (EIOS) network ...

Oct 11, 2026 AI in Africa, Machine Learning

A support note whose fields are covered by mask blocks, beside a ruler marking the character offsets of each span and a column of hashed identifiers a magnifying glass turns back into the original numbers

Redact, Then Prove It: Per-Type Recall, Offset Damage and the Reversible Pseudonym

Two numbers from this post’s fixture: the day-one regex set caught 5 of 13 PII spans, and a plain SHA-256 release gave all 24 subject IDs back to an attacker holding a 200,000-value register. Both ...

Oct 10, 2026 Machine Learning, ML Ops

A phone handset at the centre of three radiating link paths labelled mother, frontline health worker and health facility, a telecom mast with signal arcs on the right, a struck-through one-way broadcast arrow replaced by a two-way helpdesk arrow, and the targets 500,000 women, 5,000 workers and 500 facilities

The Bottleneck Is Adoption, Not Innovation: Reading Nigeria's AI Maternal Health Multiplier

Introduction The release, in one line On 22 September 2026 the Gates Foundation and the MTN Group Foundation announced the Nigeria Maternal Health Multiplier: four years, roughly US$25 million,...

Oct 9, 2026 AI in Africa, Machine Learning