These two founders left Goldman and Meta to create the market voice AI that everyone was ignoring

Customer support and service are among the hottest fields in voice AI right now. But building a product that feels human and responds without noticeable lag turns out to be more difficult in some markets than others – and most of the big players weren’t built with Africa and the Middle East in mind.

AethexAI, a startup founded last year to fill that gap, has raised $3 million in pre-seed funding led by 4DX Ventures, with participation from Enza Capital, Dorm Room Fund, Mojo Ventures, and Stanford GSB 26 Fund. Individual investors include Stanford faculty, telecom executives, and AI researchers from Anthropic.

Rather than using existing orchestration tools like Vapi and LiveKit, the company built its own micro-model and orchestration layer from scratch to handle the dialects of English, French, and Arabic spoken in all target markets – a decision driven, as it will be, by the specific operational needs of the region.

The company also launched its platform for businesses to try out its technology and sign up for its services, as well as APIs and SDKs for developers to test its models.

The startup was founded by Mariama Diallo and Ayooluwa Odemuyiwa. CEO Diallo worked at Goldman Sachs and later joined YC-backed ModelML as a product and growth hire. CTO Odemuyiwa graduated from Caltech, worked at Meta, and enrolled at Stanford Business School before co-founding the company. The two wanted to build something in emerging markets and began looking for opportunities.

Businesses around the world are rushing to use AI tools to automate certain parts of their operations. But that doesn’t always work out. In Egypt, a call center automated most of its calls, but returned the system due to negative results, the founders found. Several support centers in Africa told them that finding and hiring engineers to make calls at a reasonable cost was a constant headache.

“The delay and jitter we saw in the automated calls in this region was annoying. If we were an orchestrator, we would have to use large models that were hosted outside the region, which led to high latency. We realized that for this to work, we have to use very small models and cut the delay at every step,” Odemuyiwa told TechCrunch about the decision to build models for the company or the company layer.

AI labs running their latest models often spend millions training them and acquiring data. AethexAI has found a solution for both. Instead of chasing the biggest models, it decided that smaller models are enough to deal with the latency problem while maintaining accuracy and developed its Kora series, which has parameters ranging from 300 million to 1.7 billion. That’s part of the greatness of LLMs, which is exactly the point.

To train these models, the startup used anonymous recordings from a call center colleague. It also sent hard drives to radio stations across Africa to collect more audio data. To keep costs down, he created a network of university student contributors to interpret data and pronounce local names. Because of this, the startup says, it now handles more than 17,000 calls a day.

On the business side, the company is careful to walk new customers to express AI through the process, offering on-site demos and workshops to help them identify the best use cases for automation.

“We always tell customers that we can’t be everything to everyone right now. We’re small. When we start talking to a company, we ask them to pick one use case that’s most important to start with.” [with],” said Diallo.

The startup is open to work in all industries, but currently, the majority of its use cases involve debt collection calls, customer activation, or KYC — Know Your Customer verification, a common identity verification process used by banks and telecoms. The company hires forward-deployed developers on a contract basis to serve local markets and build channel partnerships with telecom providers to handle AI voice calls. Plug-and-play solutions, it says, won’t work here.

Walter Baddoo, founder and managing partner of 4DX Ventures, says the African and Middle Eastern market is very different from the markets most AI companies are built to serve.

“Businesses in Africa and the Middle East process almost three times the number of calls of their Western counterparts, as voice is still the leading channel for customer communication,” he said. “Operating systems built for Western markets characterized by high-end GPU infrastructure, standard English and European speech environments, and common business workflows in the US and Europe. That creates real gaps where businesses need systems that handle dialects, code switching, and informal speech patterns, and work within their existing telephony infrastructure and their real value points.”

Put another way, while companies like ElevenLabs, Deepgram, Sierra, and Cognigy are expanding globally at a rapid pace, the markets they are built for and the markets they enter are not always the same. Startups like AethexAI are betting that spaces – specialized models in local dialects, local partnerships, regionally built infrastructure – represent the opening of a market where the giants have no incentive or structures to close.