Two founders walked away from Goldman Sachs and Meta to chase a market the voice AI giants mostly ignore. According to TechCrunch AI, their startup AethexAI just raised $3 million in pre-seed funding to build voice agents tuned for Africa and the Middle East, regions where the dominant players were never really designed to operate. The round was led by 4DX Ventures, with Enza Capital, Dorm Room Fund, Mojo Ventures, and the Stanford GSB 26 Fund joining in. Some of the angel money came from AI researchers at Anthropic.
This is significant because it’s a direct bet against the assumption that one global voice AI stack fits every market.
Who’s behind it
AethexAI was founded last year by Mariama Diallo and Ayooluwa Odemuyiwa. Diallo, the CEO, worked at Goldman Sachs and later joined YC-backed ModelML in a product and growth role. Odemuyiwa, the CTO, graduated from Caltech, worked at Meta, and enrolled at Stanford Business School before the two teamed up. They wanted to build for emerging markets, so they went looking for a real problem. They found one fast.
In Egypt, a call center automated a big chunk of its calls, then rolled the whole thing back because the results were poor. Support centers across Africa told the founders the same thing: hiring engineers to automate calls at a workable cost was a constant headache, as detailed in TechCrunch AI.
Why they built their own models
Most voice AI startups stitch together existing tools like Vapi and LiveKit and call large models hosted somewhere far away. AethexAI decided that approach was the problem, not the solution.
“The latency and jitter that we saw on automated calls in this region were outrageous,” Odemuyiwa told TechCrunch. “If we had become orchestrators, we might have had to use large models that were hosted outside the region, resulting in higher latency. We realized that in order for this to work, we have to use very small models and cut latency at every step.”
So they built their own orchestration layer and a model family called Kora, ranging from 300 million to 1.7 billion parameters. That’s tiny next to the big LLMs, and that’s the point. Small models stay fast and stay local, which matters when a human is waiting on the other end of the line.
Getting training data took some unusual moves:
- They used anonymized recordings from a call center partner.
- They shipped hard drives to radio stations across Africa to collect more audio.
- They built a network of university students to annotate data and pronounce local names.
The payoff so far: AethexAI says it’s handling more than 17,000 calls a day, across localized dialects of English, French, and Arabic.
Why this matters for the industry
Walter Baddoo of 4DX Ventures frames the regional gap in hard numbers. “Enterprises in Africa and the Middle East process roughly three times the call volume of their Western counterparts, as voice is still the dominant channel for customer interaction,” he said. Incumbent systems, he argues, were built for high-end GPU infrastructure, standard English, and US and European workflows. That leaves real gaps around dialects, code-switching, informal speech, and local telephony.
What stands out here is the strategy. AethexAI isn’t trying to be everything to everyone. Diallo says they ask each new client to pick a single use case to start. Right now the work clusters around debt collection, customer activation, and KYC identity checks for banks and telecoms. The company is hiring forward-deployed engineers on contract and building channel partnerships with telecom providers to handle the actual phone calls.
The broader read: while ElevenLabs, Deepgram, Sierra, and Cognigy expand fast across the globe, the markets they were built for and the markets they’re entering aren’t the same thing. AethexAI is betting that local dialects, on-the-ground partnerships, and region-specific infrastructure add up to an opening the giants have neither the incentive nor the architecture to close.
If that bet holds, expect more region-first voice AI plays to follow, especially in markets where voice still carries most customer conversations. Full details are at the original TechCrunch AI report.