Hassan Sawaf keeps finding things that need translating

Before Hassan Sawaf worked on machines that could hold a conversation, he worked on messages about aeroplanes. They carried the practical business of arrival: passengers, cargo, baggage, fuel. Somebody at an airport had to read the shorthand, understand it and turn it into instructions. In the early 1990s, Sawaf helped build software to do that work automatically. His introduction to artificial intelligence involved the unglamorous problem of getting information to the people who needed to act on it.

That is a useful place to begin a career in a field fond of grand entrances. An airport message has somewhere to be. An interpretation has consequences. The computer must do more than produce a plausible sentence; the message has to become useful work. Across the companies Sawaf would later join, and the companies he would start, versions of that problem kept returning.

Today, he is the founder and CEO of aiXplain, based in San Jose. His career passes through Aachen, AppTek, SAIC, eBay, Amazon and Facebook. The names get larger. The question remains surprisingly intimate: how do you help someone express what they need, and make the machinery understand?

A conversation that changed the assignment

Sawaf was born and raised in Germany and has Syrian origins. German and Arabic were familiar languages long before machine translation became his work. While working on airline and airport automation at Daimler, he presented his work in 1995. Someone in the audience approached him about a German government-funded speech translation project. A professional introduction met a personal interest: the possibility of speaking in one language and being understood in another.

He studied computer science at RWTH Aachen and became a senior researcher there. The work brought together speech recognition and translation. A spoken exchange asks several things of a machine: recognise the sounds, preserve the meaning, then say something intelligible back. Anyone who has asked a voice assistant an ordinary question and received an extraordinary answer knows how many opportunities that arrangement offers for embarrassment.

The university research became a commercial venture, AIXPLAIN AG, around the turn of the millennium. In 2002, the company presented a speech-to-speech system translating between English and German. This was years before such a demonstration could be summoned casually on a phone. Selling the technology also meant dealing with the machines that would run it. Delivery could involve servers installed at a customer’s premises, rather than a convenient cloud service.

A useful idea had to survive the trip out of the laboratory. That journey would become a recurring part of Sawaf’s working life.

The shopping problem behind the language problem

AppTek acquired AIXPLAIN’s assets in 2004, and Sawaf moved to the United States. As AppTek’s COO and chief scientist, he led work on hybrid machine translation and speech recognition across multiple languages. He later became chief scientist at SAIC, working on human language technology. In 2011, the company named him a Technical Fellow.

These jobs placed language technology inside organisations with particular demands. Recognition engines had to handle different languages and applications. Dialogue systems had to support an exchange, including instructions and questions. The research had customers, constraints and a destination. Translation was becoming a working component inside a larger operation.

At eBay, which he joined in April 2013, the larger operation was commerce across borders. A seller could have precisely the thing a buyer wanted, while describing it in a language the buyer could not read. Machine translation helped make that encounter possible. Sawaf established AI teams whose work expanded into computer vision, natural language understanding and dialogue.

Shopping is an instructive test for an interface. People have a habit of wanting things they cannot name exactly. They recognise a shape, remember a colour, or know the occasion for which something is needed. A search box makes a bargain with the customer: put the right words in, and something useful may come out. Making pictures part of a chatbot conversation loosened that bargain. The customer could bring more of the problem into the exchange.

For Sawaf, the scope kept widening. A language system could help a message cross a border. It could also help an intention cross the distance between a person and a piece of software.

Very large companies. A wider audience.

Sawaf joined Amazon in September 2016, moving into AWS as director of applied science and artificial intelligence in January 2017. Work on services including Lex, Transcribe, Comprehend and Translate put language capabilities within reach of developers building their own applications. The attraction was the range of things other people might make.

“I love new ideas,” he said in 2017. He also described enjoying helping entrepreneurs get started. There is a particular pleasure in building something whose eventual use you cannot entirely predict. A reusable service leaves room for a customer to bring the idea. The customer can concentrate on a business problem while drawing on capabilities that would otherwise require a specialist team.

In October 2019, he joined Facebook AI, working on technology for augmented reality, including computer vision, signal processing, speech and language. He left in June 2020 and founded aiXplain that month. After years building inside large organisations, he wanted a platform that people outside the engineering profession could use.

The change in audience mattered. A developer can be given an API and documentation. Someone running a business may begin with an entirely different description of the task. A system can be available on the internet and still be a long way from usable. Sawaf had spent years helping computers interpret language. Now he was looking at the language separating the people building AI from the people hoping to apply it.

“So I think we need to learn to lead.”

Hassan Sawaf, on working with AI agents

A company name gets a second life

There is a small joke in the continuity. The name aiXplain reaches back to his Aachen venture. Sawaf has connected it to Aachen’s French name, Aix-la-Chapelle. When he began again in 2020, he reused the name. Customers liked its associations with AI and explanation. A name with an earlier life had found another suitable job.

The new company’s proposition was practical: let users assemble AI resources, compare them and deploy working solutions. Calibrate Ventures co-led an $8 million seed round in 2023. Its investment case centred on people who needed AI applications without possessing the technical teams to build everything themselves. A marketplace supplied models, data and other components; tools helped users combine and evaluate them.

In July 2024, aiXplain announced a $6.5 million pre-Series A led by Wa’ed Ventures. Bel Esprit, then in open beta, used natural language prompts to help nontechnical users put AI solutions together. The investment also supported expansion and localisation. The geography was part of the access question: a useful system should be able to serve people beyond the markets where its components were invented.

Funding supplies a company with time and people. The user still supplies the test. Can a person who understands the work explain it well enough to get a useful result? Can that result be checked? Those are less photogenic questions than a financing announcement, and much closer to Sawaf’s original airport assignment.

Giving the software a job description

By September 2025, Sawaf was discussing AI agents with Robert Scoble and the changing work of managers. An agent can undertake a sequence of steps towards a goal. Giving it work involves describing the goal, choosing what it can use and deciding how its output will be judged. Delegation brings an old office problem into a new technical setting: instructions that seemed perfectly clear when they left your mouth.

aiXplain now describes its mission in terms of AI that enterprises can understand, trust and own. Its headquarters are in San Jose, with a regional headquarters in Riyadh. The emphasis on accountability follows naturally from offering systems that can act. As the software acquires more room to make choices, the person responsible for the outcome needs a clearer view of those choices.

Sawaf also remains involved in published research. In March 2026, he co-authored an EACL system-demonstration paper on AI-assisted candidate assessment. The work combines job descriptions, CVs, interview transcripts and feedback into structured evaluations, using role-specific rubrics and a multi-agent design. It is a particular application with a familiar concern: make the path from messy language to a decision more explicit and inspectable.

A system that produces an answer has completed one part of the job. Someone must still decide whether the answer deserves to be used. That responsibility gives the enthusiasm for easy access its necessary weight.

There is also a world outside the screen

The personal details offer a welcome change of pace. In 2017, Sawaf’s interests included hiking, learning guitar and racing sports cars on a track in Monterey. His personal website shows him in the Narrows at Zion National Park in summer 2020, holding a walking stick beside the river. It is a cheerful image of someone whose next step requires no software specification.

His advice to founders has been less concerned with theatrical confidence than with knowing the subject, continuing to learn and finding teammates who are motivated and loyal. He has credited his Aachen supervisor, Hermann Ney, as an important mentor. He has also urged entrepreneurs to read widely and keep going when a competitor appears with a similar idea.

That seems fitting for a career spent revisiting a question in changing surroundings. Airport messages, bilingual conversations, shopping, cloud services and AI agents each put a different demand on understanding. Sawaf keeps returning to the point where a person knows what should happen and the machine needs help getting there. The vocabulary has changed considerably since the telex. The work still begins with listening.