Our research pioneers AI for low-resource African languages, from fine-tuning LLMs to building real-time voice systems.
We fine-tune translation models like Google's TranslateGemma-12B for English↔Kikuyu using Parameter-Efficient Fine-Tuning (LoRA). Our deployed model achieves 19.61 BLEU — a 758% improvement over zero-shot performance on 30,430 curated sentence pairs.
Key areas: LoRA optimization, regularization tuning, production deployment on Modal serverless GPUs
We are building end-to-end voice AI using the Mimi neural codec adapted for Kikuyu tonal fidelity. Our Stage 1 codec adaptation is complete (79.3M params, 1.1 Hz pitch error), with streaming inference and full-duplex conversation in development.
Key areas: Mimi codec adaptation, pitch-preservation loss, cascaded ASR→LLM→TTS pipeline
We use the African Next Voices corpus (750+ hours of Kikuyu audio) and Google's WAXAL TTS dataset (~9 hours studio quality) to train robust speech and translation models.
Key areas: Audio preprocessing, noise augmentation, multi-source dataset curation
We deploy models on Modal serverless GPUs (A10G, A100) with keep-warm strategies to minimize cold starts. Our Next.js frontend proxies API requests to serverless backends for production-grade serving.
Key areas: Modal serverless, keep-warm scheduling, Next.js API routes
Our technical papers on LLM fine-tuning and Speech-to-Speech architectures will be published here. Preprints coming soon.
Publications coming soon
We partner with universities, language communities, and funders to advance African language AI.
Interested in collaboration or funding our research? Contact us at hello@c-elo.com