Ex-Goldman and Meta Employees Build Voice AI for Emerging Markets

Ex-Goldman and Meta Employees Build Voice AI for Emerging Markets

Customer support and service operations are currently among the hottest areas in voice artificial intelligence (AI). However, creating a product that mimics human speech and responds without noticeable latency is a complex challenge in certain markets. Many major companies have developed their systems without considering African and Middle Eastern markets. To fill this gap, AethexAI, a startup founded last year, has raised $3 million in pre-seed funding led by 4DX Ventures. Techcrunch.com reports on this. report .

Instead of using off-the-shelf tools like Vapi and LiveKit, the company built its own small model and orchestration layer from scratch to process local dialects of English, French, and Arabic. Startup founders Mariama Diallo (ex-Goldman Sachs) and Ayooluwa Odemuyiwa (ex-Meta engineer) aimed to solve the high latency problem in automated calls within emerging markets. To achieve this, they bypassed large models hosted abroad, focusing instead on smaller models that operate locally.

AethexAI developed its Kora series models, ranging from 300 million to 1.7 billion parameters. These are significantly smaller than Large Language Models (LLMs), which is precisely what ensures speed. To train the models, the startup partnered with call centers and even shipped hard drives to collect audio data from radio stations across Africa. Additionally, a network of students was organized to ensure correct pronunciation of local names and for data labeling.

Currently, the startup successfully handles over 17,000 calls per day. The company has launched a platform for enterprises to test its technology, as well as API and SDK tools for developers. AethexAI provides not only technology but also conducts masterclasses and demos to help clients identify the best automation strategies.

Add Zamin.uz to GoogleRead "Zamin" on Telegram!
Nodirbek Razzokov
«ZAMIN.UZ» editor

Comments 0

Related news