Back to Projects
G
Data Science / Analysis

GSMA AI Talent Mapping

Data Engineer & Analyst

PythonPandasNext.jsPowerBIWeb Scraping

About the Project

A consultancy research engagement for the GSMA African AI Language Models Initiative, Talent Workstream, mapping AI and LLM talent across 5 African regions and 7 skill clusters. The pipeline collects empirical signals from sources that each see a different slice of the market — job boards (Jobberman Nigeria and Ghana, Adzuna, Remotive), GitHub repositories and contributors, and academic publications via Google Scholar, Semantic Scholar and arXiv — then merges them into a single master analysis that feeds both an interactive dashboard and the written deliverables.

What It's Made Of

Job market collectors

Jobberman NG/GH · Adzuna · Remotive

Collectors against job boards covering Nigeria, Ghana and remote listings, capturing formally advertised AI and LLM roles. Known bias, stated in the report: this source sees only formal advertised employment and misses the large informal market entirely.

GitHub signal collection

Repositories · contributors · topics · stars

Practitioner signals from GitHub — repositories, contributors, topics and stars — as evidence of people actually building rather than roles being advertised. Also biased in a knowable direction: it over-represents open-source contributors and under-represents enterprise engineers who ship nothing publicly.

Academic publication collection

Google Scholar · Semantic Scholar · arXiv

Research output across three academic search sources, mapping where AI and LLM work is being published on the continent. Skews toward academia over industry, which is why it is one of three views rather than the answer.

Master analysis & rebuild

Python · Pandas · uv

The reconciliation step that merges all three sources into one master dataset through a repeatable rebuild rather than manual merging, segmented across five African regions and seven skill clusters. Each source is weighted as a complementary partial view, because treating any one as ground truth would produce a confidently wrong map.

Talent dashboard

Next.js · static export

An interactive dashboard shipped as a statically exported Next.js app, deployed both at a root domain and under a path prefix so it could be embedded in the client's own site as well as stand alone.

Written deliverables

Report · executive brief · interview guides

The engagement's actual output: a Talent Landscape Mapping and Pathways report, an executive brief, interview guides, and analysis of survey responses — all drawing on the same master dataset the dashboard renders, so the narrative and the numbers cannot diverge.

Key Highlights

  • Built collectors across job boards, GitHub, and academic search, reconciling them into one master dataset via a repeatable rebuild step rather than manual merging
  • Analysed AI talent across 5 African regions and 7 skill clusters, using GitHub topics, contributors and stars as practitioner signals alongside formal job demand
  • Shipped the talent dashboard as a statically exported Next.js app, deployed both at a root domain and under a path prefix
  • Produced the engagement's deliverables — a Talent Landscape Mapping and Pathways report, an executive brief, interview guides, and survey-response analysis
  • Managed the Python toolchain with uv for reproducible collector runs across the project's lifetime

Technical Challenges

Aggregating talent signals from completely different sources into a unified view required careful normalization, because each source is biased in a knowable direction: GitHub over-represents open-source contributors and under-represents enterprise engineers, job APIs capture only formal advertised employment and miss the large informal market, and publication data skews toward academia over industry. Treating any one of them as ground truth would have produced a confidently wrong map, so the analysis weights them as complementary partial views and says so explicitly in the report.