Models & tools
Try Somali language and speech models in the browser.
Soplang — the first Somali programming language
Built by our team, Soplang lets Somali speakers write real software using native Somali keywords instead of English syntax — a compiler, not just a translation layer.
224
GitHub stars
64
Forks
v2.0.0
Latest release
MIT
License
// Salaan, Adduunka! — Hello, World!
qor("Salaan, Adduunka!")
door magac = "Sharafdin"
abn age = 25
hawl salaam(qof) {
celi "Salaan, " + qof + "!"
}
qor(salaam(magac))Somali-based syntax
Keywords and concepts written in Somali — "door", "qor", "hawl" — not transliterated English.
Dual type system
Static typing ("abn", "qoraal") alongside dynamic typing ("door") in one language.
Compiled execution
A Cranelift JIT for running files directly, plus an ahead-of-time build path for standalone native binaries.
Modern tooling
CLI, REPL, and full object-oriented support — classes, inheritance, encapsulation.
The domains we work across
A full-stack research practice spanning language, speech, and retrieval — built for low-resource settings.
NLP
Tagging, parsing, NER
ASR
Speech to text
TTS
Text to speech
LLMs
Language models
Tokenizers
Subword for Somali
Benchmarks
Task suites
Evaluation
Rigorous metrics
RAG
Grounded retrieval
Speech Processing
Alignment & diarization
Frequently asked
How accurate is whisper-som-small?
It reaches 11.4% word error rate on held-out Somali radio audio, trained on 320 hours of aligned speech with word-level timestamps.
How much more efficient is goobo-tokenizer than general-purpose tokenizers?
It averages 1.53 tokens per Somali word, versus 2.69 for BERT-base — a 1.75× improvement that means shorter sequences and cheaper training and inference for downstream Somali models.
What is Soplang, and is it related to the AI models?
Soplang is a separate, MIT-licensed Somali-inspired programming language — not an AI model. It's part of the same open-source effort to build technology in Somali, not just for Somali speakers.
Build Somali AI in the open
Use our datasets and models, contribute to the research, or partner with the lab. Everything we can open, we do.