Once again, artificial intelligence dominated the buzz at this year’s MWC Barcelona, formerly called Mobile World Congress. From smartphones to satellites, networks to applications, no vendor or service provider could resist the lure of AI.

My challenge: Sort through the thousands of exhibitors to uncover stories of innovation that might otherwise get lost amidst the hubbub. From the 2,900 exhibitors overall to the 1,000+ touting AI innovation, I pared down the list to 30 interviews.

Of these, nine made the cut. Here are my choices:

Yeo Messaging: WhatsApp on steroids

WhatsApp, Telegram and other messaging applications tout their end-to-end secure encryption that protects the privacy of their users. What they don’t offer, however, is a way to confirm that an intended user is the person actually communicating.

Yeo Messaging Ltd. solves this problem by verifying the face of the user in real-time. If the app doesn’t detect the right face, it blurs out the screen in about half a second.

Turn the phone away, it blurs. Turn it back, the message reveals itself. Yeo also offers a software development kit that enables third parties to integrate the company’s continuous facial recognition into their own apps.

SconAI: Simultaneous language interpretation

Language interpretation (real-time translation of voice to text) is a tough market, given the ubiquity of Google Translate.

SconAI Co. Ltd. delivers the goods. It interprets human voice audio in real-time, displaying the translated text as the human speaks.

Instead of waiting until the end of a sentence, SconAI makes its best guess as a sentence progresses, then revising in real-time as the speaker completes the sentence. In this way SconAI optimizes the interpreted output without introducing a delay.

Hypergraph: Using graph neural networks for network intrusion detection

Today, every cybersecurity vendor under the sun is integrating large language models or LLMs into their products. In addition, vendors have been using machine learning to uncover anomalies in large data sets for years, including indicators of compromise in cybersecurity scenarios.

Hypergraph AI takes a different approach. The company uses graph neural networks – neural networks that take graph-formatted data as input – to detect deviations from normal network behavior.

Graphs are the natural way to represent networks, so inputting graphs into AI models is an obvious approach. However, by using neural networks, Hypergraph can represent graphs of networks as they change over time.

This temporal component makes all the difference, as it both picks up changes and uncovers root causes, leading to better detection and fewer false positives as compared with competing products on the market.

Fenix DFA: Confirming your backups are actually working

How do you know your backups are working? You wait for a disaster, attempt to recover data from backup, and cross your fingers. Clearly, this finger-crossing is not good enough – especially when regulations mandate specific levels of data resilience.

Fenix DFA (a KYMO Investments Inc. portfolio company) addresses this problem with proactive data resilience via a governance layer for backup operations.

This governance layer confirms the backups are working and in general, that they comply with relevant policies.

Fenix uses behavior analysis to uncover hidden backup risks, identifying issues that ops teams don’t normally track. It then provides workflows for addressing any issues. It even delivers continuous auditing of data resilience processes for compliance purposes.

Enhans: Using ontologies and rules to eliminate hallucinations in AI-driven workflows

Many vendors are leveraging LLMs to offer AI agents that can both execute tasks individually and coordinate workflows that orchestrate agentic actions. As with other generative AI applications of LLMs, however, inaccurate results, aka hallucinations, are the Achilles heel of such agentic workflows.

Enhans Co., Ltd. solves the problem of hallucinations by leveraging a proprietary ontology – a cross-referenced vocabulary for the type of workflow in question.

It then combines agentic behavior with a traditional rules engine to deliver domain-specific orchestrated workflows without hallucinations.

Ecommerce is Enhans’ starting point. The company transforms complex e-commerce processes with real-time market intelligence, its proprietary ontology and its agentic platform.

Diagnext: Using AI to build a better data compression mousetrap

Data compression has been around for decades. I’m so old that I remember zipping files back in the 1980s.

Data compression formats such as JPEG and MPEG for images, audio and video are well-established standards we all take for granted. Given AI’s unquenchable thirst for massive data files, an improved approach to data compression can improve the performance of AI-based applications while reducing the cost of their operations.

To address these needs, Diagnext.com SA takes data compression to the next level with an approach it calls adaptive compression. Using AI-based analysis of files and other data, the company can achieve dramatically better compression ratios than traditional compression approaches, without the loss of any useful data. Diagnext can even compress previously compressed files, for example, reducing the size of MPEG files by up to 82%.

Diagnext’s compression is especially useful for industry-specific data formats. For example, it can substantially reduce the size of massive genomics files with no loss of data. It can even compress x-ray image files with results that comply with standard medical data formats with no loss of diagnostic capability. Hospitals, in fact, are among Diagnext’s early adopters.

Cubig: Multimodal, AI-ready synthetic data

Privacy considerations typically prevent organizations from using production data for testing or training their AI models.

Many vendors offer synthetic data that is similar enough to production data to fill these needs. However, synthetic data is usually text-based. A common example: customer or patient records with names and other personally identifiable information replaced with similar but fictitious values.

Today’s AI, however, deals with more than text. What about images or other types of information, aka multimodal data formats?

Cubig Corp. can generate synthetic multimodal data sets. For example, the company can create synthetic x-rays that are sufficiently realistic to support application testing and model training purposes.

Military applications include replacing the backgrounds in images of drones to yield synthetic data sets that do not include any information about the location of each photo. Cubig can also generate specialized text-based synthetic data sets as well, for example, simulated survey responses.

Calix: Agentic platform for providing value-added services to telco subscribers

Agentic platforms that purport to leverage AI agents to automate workflows are springing up like weeds. Calix Inc., however, stands out from the crowd. It offers an agentic platform for communications service providers or CSPs – a fancy term for the company that provides internet, phone and television services to consumers and businesses.

Not only do today’s CSPs offer a complex suite of services, but pricing pressure in the market is forcing them to provide superior service as well as value-added services to remain competitive. Calix offers such transformation to these service providers. CSPs can rebrand the Calix platform and use it to support the full lifecycle of subscriber relationships, from onboarding and operations to cross-selling and upselling value-added services.

Calix also offers platforms for CSPs that target small businesses as well as outdoor applications such as campuses and stadiums.

Arrcus: Policy-driven network for distributed inferencing

Arrcus Inc. is an established player in the networking space with its distributed, policy-based network fabric offering. The company’s new offering, however, is sufficiently innovative to make my short list.

Leveraging its networking fabric, Arrcus now supports distributed inferencing, where AI-based applications make decisions and take action at various points across a distributed network.

Take, for example, a modern factory floor, with various machines, cameras, vehicles, and other devices – all of which leverage AI to interpret data and in many cases, take particular actions. Such factories want to push the necessary inferencing to these endpoints to reduce bandwidth consumption and avoid latency-related performance issues.

With Arrcus, these companies now have a policy-based network that can support various endpoints with distributed inferencing, enabling them to control security, performance and other network capabilities in real time across the network.

The Intellyx take

When you set out to look for innovation, you are likely to find it in various unexpected corners rather than directly on your path. Such is the nature of innovation: It is always surprising.

For an innovative vendor to take the leap into exhibiting at a massive conference like MWC, furthermore, means that they have more than some surprising new widget. They must also be convinced that there’s a market for that widget – some customer pain point they are uniquely qualified to address.

The nine vendors in this article have made this leap. Some may succeed while others may fail, to be sure – but such risks are also an integral part of innovation. And without such risks, there’s no way to achieve the commensurate rewards.

Jason Bloomberg is founder and managing director of Intellyx, which advises business leaders and technology vendors on their digital transformation strategies. He wrote this article for SiliconANGLE. Disclosure: Arrcus is a former Intellyx customer.

Photo: MWC