Phase 3 · February 2026

Roadmap & language coverage

Track the milestones and language tracks we are funding. See how the February 2026 launch is shaping up.

0d 0h until launch

Language tracks

Nguni family

isiZulu, isiXhosa

Fine-tunes with grounded retrieval corpora sourced from curriculum and civic data.

65% complete

Sotho–Tswana family

Sesotho, Setswana

Datasets curated with provincial partners and universities.

45% complete

Swahili & Afrikaans

Kiswahili, Afrikaans

LoRA adapters plus translation benchmarks aligned with SA financial terms.

30% complete

Dataset & checkpoint pipeline

Acquire

Done

POPIA-compliant ingestion from government gazettes, educational resources, and open civic datasets.

Cleanse

Done

Deduplicate, redact, and score documents before they enter the contributor task queue.

Align

In progress

Instruct-tune checkpoints on a mix of open-source and MafutaAI-owned corpora.

Evaluate

Upcoming

Run multilingual eval suites to promote checkpoints to the public catalog.

Weekly focus

  • Finalize VAT and municipal retrieval datasets.
  • Prototype Nguni-family LoRA adapters on Cape Town racks.
  • Scale contributor verification tooling for remote GPU onboarding.

Weekly items reflect current engineering sprints. They will expand as we approach the public beta.