About Acougloss
The problem
African call centres and fintech CX teams run on Kiswahili — often mixed with English, Sheng, and mother-tongue accents. Mainstream speech recognition is trained largely on other domains and collapses on the turns that matter: amounts, dates, product names, and clarifications.
What we do
We build domain-adapted ASR for African languages, starting with Kenyan Kiswahili CX. We fine-tune from strong open baselines (Whisper), evaluate with transparent word error rate, and ship a production API with side-by-side compare mode so you can see improvement on your own audio before you roll out.
Where we are
Live production system and public demo at acougloss.ai. Our published checkpoint shows large gains on a public Kiswahili read-speech benchmark versus out-of-the-box Whisper — proof the adaptation loop works. We are actively collecting and partnering for real Kenya telephony and code-switched CX audio so pilots can be measured on the audio that matters. Accepting pilot customers in CX, BPO, and financial services.
The team
We are a team of enthusiasts with expertise in AI, data and business, passionate about accessibility of AI, and speech recognition in particular, in low-resource language markets.
Contact
Nikolay & Team