About
An index built in the open
goha (ጎሃ — “dawn” in Amharic) exists because the Ethiopian AI ecosystem was visible everywhere and legible nowhere. Models were scattered across Hugging Face, datasets across GitHub, research across arXiv — and nobody could answer a simple question: what exists for Amharic, Afaan Oromo, Tigrinya and Ge'ez, and what is missing?
This site is that answer, rebuilt every day: a live index of models, datasets, research, organizations and benchmarks — with the gaps stated as plainly as the wins.
How it works
- 1
Scrape the public record
Every day, goha queries Hugging Face, GitHub and arXiv for models, datasets, spaces, repositories and papers across Amharic, Afaan Oromo, Tigrinya, Ge'ez and more.
- 2
Normalize into one contract
Everything becomes a single Project shape — title, owner, task, languages, downloads, license, model card, evaluations, lineage. The same record powers every page.
- 3
Derive the ecosystem surfaces
Activity feed, movers, organization footprints, research tags, leaderboards, language coverage, growth history and the family tree are all computed from the same scrape.
- 4
Export a static site
The build emits plain JSON and HTML. No database, no server, no API keys — the index is files, deployed to GitHub Pages and refreshed by the daily commit.
Principles
Every number links to its source
Benchmarks cite the model card they came from. Nothing is estimated, inferred beyond published metadata, or invented to look complete.
Amharic is not decoration
Languages, scripts and local names are first-class data — coverage and gaps are reported per language, not as an aggregate.
The scraper is the product
The interface is a surface over the pipeline. If a number is wrong here, the fix belongs in the scrape, not in a hardcoded label.
No LLM required
Ask Goha answers from structured data with a deterministic parser. It is inspectable, free and never fabricates an answer.
What it deliberately does not do
- It does not host or run models — every link goes to the source (Hugging Face, GitHub, arXiv).
- It does not claim completeness: the index only knows what is published with discoverable metadata.
- It does not rank by opinion — leaderboards only aggregate numbers the authors published.
- It does not generate text — Ask Goha composes answers from the same structured data you can read in the open data.
Who built this
goha is built by Henok Tekeba — a student who wanted this index to exist, so he built it. The scraper, the data pipeline and the ecosystem surfaces are original work; the interface is adapted from OpenSource Together (CC BY-NC 4.0).
Source and daily snapshots: github.com/Henok-Tekeba/goha · Data: MIT · UI: CC BY-NC 4.0
Explore: lineage · benchmarks · gaps · ask · open data
