Title: Ex-Goldman and Meta Executives Launch Voice AI Startup Targeting Underserved African and Middle Eastern Markets

While customer service represents a booming sector for voice AI, developing technology that mimics human speech with minimal delay proves significantly more complex in certain regions. Historically, major industry players have overlooked the specific needs of Africa and the Middle East. Addressing this gap, AethexAI, a startup established last year, has secured $3 million in pre-seed funding. The round was led by 4DX Ventures, with additional investment from Enza Capital, Dorm Room Fund, Mojo Ventures, and the Stanford GSB 26 Fund. The investor list also features individual backers, including Anthropic AI researchers, telecom executives, and Stanford faculty members.

Rather than relying on standard orchestration platforms such as Vapi or LiveKit, AethexAI engineered its own orchestration layer and small-scale models from the ground up. This strategic choice allows the company to manage the localized dialects of English, French, and Arabic prevalent in its target regions. Concurrently, the startup is rolling out its enterprise platform for service subscriptions, while releasing APIs and SDKs to allow developers to test its models.

The company was co-founded by CEO Mariama Diallo and CTO Ayooluwa Odemuyiwa. Diallo previously worked at Goldman Sachs before joining the YC-backed firm ModelML to lead product and growth initiatives. Odemuyiwa, a Caltech graduate, worked at Meta and attended Stanford Business School before co-founding AethexAI. The duo identified a critical need in emerging markets after observing that global AI adoption often fails to translate effectively. For instance, they discovered that an Egyptian call center reverted to manual operations after an automated system yielded poor results. Similarly, several African support centers reported that hiring engineers to automate calls at a viable cost remained a persistent challenge.

“We saw outrageous levels of latency and jitter in automated calls within this region,” Odemuyiwa explained to TechCrunch regarding their technical approach. “If we had simply acted as orchestrators, we would likely have relied on large models hosted outside the region, which would have exacerbated latency issues. We realized that success required using very small models and minimizing latency at every stage.”

Unlike AI laboratories that invest millions in training massive models and acquiring vast datasets, AethexAI adopted a more efficient strategy. The founders determined that smaller models could resolve latency issues without sacrificing accuracy. They developed the "Kora" series, featuring models with parameters between 300 million and 1.7 billion—significantly smaller than typical large language models (LLMs). To train these models, the startup utilized anonymized recordings from a partner call center. Additionally, they distributed hard drives to radio stations across Africa to gather more audio data. To control expenses, AethexAI recruited a network of university students to annotate data and pronounce local names correctly. Consequently, the startup now processes over 17,000 calls daily.

On the commercial front, AethexAI assists clients new to voice AI through onsite demonstrations and workshops, helping them identify optimal automation use cases. “We are transparent with clients that we cannot serve everyone at once,” said Diallo. “We are a small team. When engaging with a new company, we ask them to select one primary use case to begin with.”

Although the startup is open to various industries, its current primary applications include debt collection, customer activation, and KYC (Know Your Customer) verification, which is the standard identity-checking procedure for banks and telecom providers. The company is currently hiring forward-deployed engineers on a contract basis to support local markets and is building channel partnerships.

Source: TechCrunch Generated at: 2026-06-03 15:00:00 UTC