Former Goldman Sachs and Meta executives launch voice AI startup targeting underserved industries

Voice AI has rapidly become one of the most active segments within customer support and service technology. However, creating systems that sound natural and respond with minimal delay remains a significant challenge, particularly in regions where language diversity, infrastructure constraints, and localised speech patterns differ from those of the markets most AI providers were originally designed to serve.

AethexAI, a startup established last year to address those challenges across Africa and the Middle East, has raised $3 million in pre-seed financing. The funding round was led by 4DX Ventures and included participation from Enza Capital, Dorm Room Fund, Mojo Ventures, and Stanford GSB 26 Fund. The investor group also includes Stanford faculty members, telecommunications executives, and artificial intelligence researchers affiliated with Anthropic.

Rather than relying on established orchestration platforms such as Vapi or LiveKit, AethexAI chose to develop its own proprietary models and orchestration infrastructure. The company says its technology was specifically designed to support localised dialects of English, French, and Arabic commonly spoken throughout its target regions.

According to the founders, building a dedicated technology stack was necessary to address the unique operational challenges associated with deploying voice AI in these markets.

Alongside the funding announcement, the company is launching its enterprise platform, enabling businesses to evaluate its technology and begin using its services. The startup is also introducing APIs and software development kits (SDKs) that allow developers to experiment with its models and integrate them into their own applications.

Mariama Diallo and Ayooluwa Odemuyiwa founded AethexAI. Diallo, who serves as CEO, previously worked at Goldman Sachs before joining Y Combinator-backed ModelML, where she focused on product and growth initiatives. Odemuyiwa, the company’s CTO, graduated from Caltech, worked at Meta, and later enrolled at Stanford Graduate School of Business before co-founding the startup.

The founders shared a common interest in building technology solutions for emerging markets and began searching for opportunities where artificial intelligence could address unmet needs.

As businesses worldwide increasingly adopt AI tools to automate operations, implementation challenges often emerge. During their research, the founders discovered organisations that struggled to achieve satisfactory results with existing voice automation solutions. In Egypt, for example, a call centre that automated a substantial portion of its customer interactions eventually scaled back deployment after experiencing performance issues. Across Africa, several support centres reported ongoing difficulties finding engineers capable of building and maintaining voice automation systems at sustainable costs.

“The latency and jitter that we saw on automated calls in this region were outrageous,” Odemuyiwa said while discussing the decision to create proprietary models and infrastructure. “If we had become orchestrators, we might have had to use large models that were hosted outside the region, resulting in higher latency. We realised that for this to work, we have to use very small models and cut latency at every step.”

Training advanced AI models typically requires substantial financial resources and access to large volumes of specialised data. AethexAI approached both challenges differently. Rather than competing to build increasingly large language models, the company focused on developing smaller models optimised for speed, responsiveness, and localised performance.

Its proprietary Kora model family ranges from 300 million to 1.7 billion parameters, significantly smaller than many mainstream large language models. The company argues that this smaller size is essential for minimising latency while maintaining practical accuracy in voice interactions.

To train these systems, AethexAI utilised anonymised recordings provided by a call centre partner. The company also distributed hard drives to radio stations across multiple African countries to collect additional speech data. To further expand its datasets while managing costs, the startup created a contributor network consisting of university students who assist with data annotation and pronunciation of local names and regional speech patterns.

As a result of these efforts, the company says its systems are now processing more than 17,000 calls every day.

Beyond technology development, AethexAI is focusing on helping businesses successfully adopt voice AI. We work closely with customers through onsite demonstrations and workshops to help organisations identify automation opportunities that deliver immediate value.

“We always tell customers that we cannot be everything for everybody right now. We’re small,” Diallo said. “When we start talking to a company, we ask them to pick one use case that is the most important to them to start with.”

While the company remains open to serving organisations across multiple industries, many of its current deployments focus on use cases such as debt collection, customer onboarding, and Know Your Customer (KYC) verification for banks, telecommunications providers, and other financial institutions.

To support regional expansion, AethexAI is hiring forward-deployed engineers on a contract basis to work directly within local markets. The company is also establishing channel partnerships with telecommunications providers to manage voice connectivity and telephony infrastructure required for AI-powered calling systems.

According to the founders, off-the-shelf solutions developed for Western markets often fail to meet the realities of local operating environments.

Walter Baddoo, co-founder and managing partner of 4DX Ventures, believes enterprises across Africa and the Middle East face challenges that differ significantly from those addressed by most existing voice AI providers.

“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,” Baddoo said. “Incumbent systems were built for Western markets characterised by high-end GPU infrastructure, standard English and European speech environments, and enterprise workflows common in the U.S. and Europe.”

He added that organisations in these regions often require systems capable of handling dialects, code-switching between languages, informal speech patterns, existing telecommunications infrastructure, and pricing models that align with local economic realities.

As global voice AI companies such as ElevenLabs, Deepgram, Sierra, and Cognigy continue to expand internationally, AethexAI is betting that significant opportunities remain in markets where localisation, infrastructure constraints, and regional partnerships create barriers that larger providers may struggle to overcome effectively.

By focusing on specialised models, regional speech patterns, and locally tailored deployment strategies, the startup aims to carve out a position in a rapidly growing market that many global competitors have yet to serve fully.