This article first appeared in Digital Edge, The Edge Malaysia Weekly on August 10, 2026 - August 16, 2026

The promise of artificial intelligence (AI) in economic development runs into the same wall everywhere where the models are only as good as the data fed into them and, in most industries, that data is either incomplete, legally restricted or simply non-existent. Athenatech.ai, a Malaysian start-up founded in 2024, is building its business in hopes of bridging that gap.

The company’s core product is a synthetic data platform called AxxonAI that generates artificial datasets designed to behave like real data, producing the same patterns, distributions and statistical relationships, without containing any actual personal information.

“Synthetic data is basically working to bypass this particular issue of real-time data. Real-time data means personal data. We are not going there,” says Sonny Dey, co-founder and CEO of Athenatech.

The distinction matters because it is precisely what makes the technology viable in regulated industries, says Sonny.

Under Malaysia’s Personal Data Protection Act 2010 and equivalent frameworks overseas, sharing patient records, financial histories or customer profiles with a third party for modelling or analysis is either heavily restricted or prohibited outright.

A client provides Athenatech with demographic profiles, behavioural parameters or aggregated patterns, none of which constitute personal data under existing privacy law, and AxxonAI generates a statistically coherent dataset that can then be used for modelling, forecasting or AI training.

The platform launched its first version in February this year. In early pilots, the synthetic output has been benchmarked against real data and recorded accuracy rates above 85%, though Sonny is careful to note that results vary by project and testing is ongoing.

Healthcare is where the technology has found its clearest application and where its limitations are also most visible, says Sonny.

In one pilot, a hospital in Petaling Jaya, Selangor, wanted to identify which patient segments in Sections 16 and 17 were most at risk for diabetes and how to direct its specialist resources accordingly. Bound by patient confidentiality rules, it could not share its records.

Instead, it provided age brackets, gender ratios, socioeconomic profiles and visit frequency data. From those parameters alone, Athenatech generated a synthetic dataset of roughly 5,000 to 6,000 patient profiles.

When the hospital benchmarked that output against its own internal records, the match came in above 85%, Sonny says.

A separate project is focused on cervical and ovarian cancer, modelling family tree data drawn from genome sequencing to generate a probability index of disease risk across generations. That project is at an earlier stage and Sonny is measured about its current state.

“The accuracy is moving gradually as we test but we are not at 90% yet [in certain models].”

Agriculture represents a different kind of data problem, says Sonny. Where healthcare data exists but cannot legally be shared, he explains that agricultural data in Malaysia is sparse and unstructured, with rubber and palm oil plantation yield records, climate impact figures and farmer productivity data held across disconnected systems with little standardisation.

Sonny reveals that the National AI Office has asked Athenatech to work on modelling crop yield patterns and the economic impact of climate events on plantation output, a mandate that remains in early development.

The SME application is the most commercially immediate, if also the most modest in scale. Sonny describes working with small food and beverage operators who want to assess whether a second outlet is viable but have no analytical infrastructure to make that call.

The platform takes in footfall estimates, catchment area demographics, weather patterns and day-of-week trading profiles and produces a revenue forecast.

The pricing runs on a pay-per-use basis, deliberately kept accessible to operators who cannot commit to large upfront costs.

“Today, it is a restaurant but it can be anything ... a person supplying spare parts to automotive companies. We want to sustain their revenue,” Sonny says.

The tool is currently being tested with a small number of operators and has not been formally launched.

Athenatech is working towards ISO standardisation for the Malaysian market and is in early discussions with global AI providers who want localised Asian datasets. Sonny argues that the major platforms cannot adequately serve the region on their own, given the linguistic and cultural complexity involved.

“Every OpenAI or ChatGPT model has inbuilt synthetic data but that data is attuned to AI training in a Western context.

“They don’t have the cultural similarity that we are building here. We are talking about Bahasa, Mandarin, Tamil. Asia is a very complex structure. In India, there are 48 languages alone; in Malaysia, there are so many. That is where we can help.”

Save by subscribing to us for your print and/or digital copy.

P/S: The Edge is also available on Apple's App Store and Android's Google Play.