
SautiLM is a research-driven startup. We build small open models that understand and speak the local languages of Eastern DR Congo, without Internet, directly on a phone.
AI tools speak French, English and standard Swahili. The languages spoken at home in Ituri and Kivu are simply not there.
In the conflict zones of Eastern DRC, displaced children learn in a language they do not master. No educational tool exists in theirs, such as Bbadha.
Our models run on entry-level phones and solar-powered terminals: no network, no data costs, and nothing leaves the device.
A foundation model series (Jua, Nuru, Wazo, Moto, Kuu, Mali) powering 10 specialized domain models and an offline mobile SDK.
A reading, general science and mother-tongue learning tutor operating without connectivity.
Multilingual translation and language comprehension bridge (Bbadha, Kingwana Swahili, French, English).
Verified arithmetic problem-solving, algebra, percentages and commercial calculations.
Public health triage, hygiene protocols, first aid and patient-caregiver translation.
Offline legal assistant grounded in DR Congo law (Constitution, Family Code, Penal Code, Labor and OHADA).
Offline camera OCR for written exercises or documents and visual field diagnostics.
Agronomic diagnosis for cassava, maize, coffee and beans, soil fertility and seasonal guidance.
Offline speech recognition (ASR) and tonal speech synthesis (TTS) for Eastern DRC languages.
SME accounting, village savings associations (VSLA/AVEC) and financial calculations in FC and USD.
Civil status documentation, public administration guidance and community mediation.
A software kit to embed our offline models directly into your own mobile applications.
Our models stay open. We earn from the applications, services and support built around them.
Digitization of bilingual dictionaries (starting with Bbadha-Swahili) and acoustic data collection via an ethical participatory approach.
Specializing and distilling open models (Gemma, wav2vec 2.0, mBART-50) for extremely low-resource languages.
Post-training model quantization (INT8) for low-latency offline execution on mobile devices.
Our foundation model series (Jua, Nuru, Wazo, Moto, Kuu, Mali), distillation pipelines and mobile testing gallery are open to researchers, developers and partner organizations.
First model in our foundation series (Jua, Nuru, Wazo, Moto, Kuu, Mali), specialized via LoRA SFT on 23,171 instructions (Bbadha, Swahili, French, English) and packaged in official LiteRT-LM INT8 format.
The open-source corpus engineering, multilingual instruction synthesis, GPU orchestration and mobile Edge-AI export pipeline.
Scientific evaluation suite (chrF++, BLEU-4 and tonal/implosive accuracy) measuring our models across Bbadha and all 10 specialized domains.
SautiLM is a research-driven startup founded by Georges Byona, Abraham Imani & Jacques Uwonda. It brings together researchers, engineers, linguists and humanitarian experts from Eastern DR Congo.
We are looking for field partners for our first pilots, academic collaborations, funding to build the corpora, and organizations that want to reach people in their own language.
hello@sautilm.com