These two founders left Goldman and Meta to build voice AI for markets everyone else overlooked

In the rapidly evolving landscape of voice AI, the focus on customer support has intensified. However, creating a solution that sounds human and responds promptly is significantly more challenging in certain regions, particularly Africa and the Middle East, which have largely been overlooked by major players. To address this gap, AethexAI, a startup founded last year, has successfully secured $3 million in pre-seed funding, spearheaded by 4DX Ventures, with additional backing from Enza Capital, Dorm Room Fund, Mojo Ventures, and Stanford GSB 26 Fund. Notable individual investors include faculty from Stanford, telecom leaders, and AI experts from Anthropic. Unlike many companies that rely on existing orchestration tools like Vapi and LiveKit, AethexAI opted to develop its own compact model and orchestration layer from the ground up. This innovative approach was necessary to accommodate the diverse dialects of English, French, and Arabic prevalent in their target regions. The founders are also set to launch a platform that allows enterprises to trial their technology, along with APIs and SDKs for developers eager to experiment with their models. The brains behind AethexAI are Mariama Diallo and Ayooluwa Odemuyiwa. Diallo, who serves as CEO, previously worked at Goldman Sachs before joining YC-backed ModelML in a product and growth role. Odemuyiwa, the CTO, is a Caltech graduate and former Meta employee, currently pursuing his studies at Stanford Business School. Driven by a desire to create solutions for emerging markets, they identified opportunities within the voice AI sector. As businesses globally rush to implement AI tools for process automation, challenges abound. For instance, a call center in Egypt faced difficulties when it automated a significant portion of its calls, eventually reverting to traditional methods due to subpar outcomes. Many support centers across Africa have noted that sourcing engineers capable of automating calls at a reasonable cost remains a constant struggle. Odemuyiwa explained the reasoning behind their decision to create in-house models and orchestration layers: "The latency and jitter we experienced with automated calls in this region were unacceptable. If we had relied on large models hosted outside the region, we would have faced even greater latency issues. We realized that using smaller models was essential to minimize latency at every stage." AI labs typically invest millions in training their latest models and gathering data. AethexAI has found a more effective route. Instead of pursuing large-scale models, they focus on smaller ones that effectively address latency while ensuring accuracy. Their Kora series, with parameters ranging from 300 million to 1.7 billion, reflects this strategy, being significantly smaller than many large language models (LLMs). To train these models, AethexAI has utilized anonymized recordings from a partner call center and dispatched hard drives to radio stations across Africa to gather additional audio data. The company has also established a contributor network of university students to assist in data annotation and the correct pronunciation of local names. As a result, AethexAI is now managing over 17,000 calls daily. On the business front, AethexAI is committed to guiding clients unfamiliar with voice AI through the implementation process. They offer on-site demonstrations and workshops to help organizations pinpoint the most critical use cases for automation. Diallo emphasized, "We always tell customers that we cannot cater to everyone right now. We’re a small team. When engaging with a company, we encourage them to identify one key use case to start with." Although the startup is open to various industries, a significant portion of its current applications involves calls related to debt collection, customer activation, and KYC (Know Your Customer) verification, which is a standard identity-checking procedure utilized by banks and telecom companies. To enhance local market engagement, AethexAI is hiring contract engineers and forging partnerships with telecom providers to manage telephony for their voice AI services. Walter Badoo, co-founder and managing partner of 4DX Ventures, highlighted the unique characteristics of the Africa and Middle East market. He stated, "Enterprises in these regions handle approximately three times the call volume of their Western counterparts since voice remains the primary channel for customer interaction. Existing systems were designed for Western markets with advanced GPU infrastructure and standard speech environments, creating significant gaps for enterprises needing systems that accommodate dialects and informal speech while fitting into their existing telephony frameworks and budgetary constraints." In contrast to companies like ElevenLabs and Deepgram, which are expanding globally, AethexAI aims to capitalize on these gaps by offering specialized models and localized partnerships, addressing the needs that larger companies may overlook.