According to Australian government figures, 1,326 people died on Australian roads in the 12 months to July 2026, up 0.9% from the previous 12-month period. The latest available national data also shows that road crashes led to around 36,000 hospitalised injuries. Yet many near misses are never reported.

Consider a driver turning across a pedestrian crossing without noticing someone still on the road. They miss each other by a few seconds. No one is injured, and the incident is soon forgotten. For transport authorities, however, that near miss could be an early warning of a dangerous intersection.  

Australian intelligent transport startup XVision AI wants to help road authorities capture these warning signs and respond earlier. “We’re trying to create the intelligence layer for roads and infrastructure,” founder Simon Maselli says. His goal is to build smart road infrastructure that can see what is happening, understand the risks and take action when necessary.

Founded in 2023, XVision AI develops AI-powered traffic management systems for governments, transport agencies and infrastructure operators. Its technology monitors how vehicles, cyclists and pedestrians move through intersections, roads and work zones. It can count and classify road users, measure traffic speeds and queues and identify potentially dangerous interactions in real time.

Maselli came to road technology through engineering, industrial systems and testing. He previously worked on complex projects for companies including BHP Group, a major Australian multinational mining and metals corporation. He also spent time at Keppel FELS in Singapore, where he was involved in certification, approvals and scientific testing for ships.

The idea for XVision came after Maselli started his own testing company about ten years ago. One customer relied on an infrastructure system assembled from components made by several vendors. When the original provider failed to service it properly, the customer asked Maselli’s company to reproduce it.

His team succeeded, but the replacement inherited the same fundamental problem: it was large, complicated and made up of too many separate components. Maselli saw a similar pattern in road infrastructure, where one product might handle detection, another communication and additional equipment that allows for the connections between them.  

“I thought there had to be a smarter way to put all those parts into one small device,” Maselli recalls. The solution also had to be easy to install without extensive roadworks, expensive components or several suppliers. That thinking eventually led to EagleEye, XVision AI’s flagship road intelligence system.  

EagleEye uses two camera lenses to create a three-dimensional view of the road. Unlike an ordinary camera that only records footage, the device uses edge AI to process what it sees locally. It detects road users, follows their trajectories and analyses how they interact.

Maselli describes its functions in three stages: perception, thinking and communication. The system first identifies vehicles, pedestrians and cyclists. It then examines their speed, direction and behaviour. Finally, it can communicate with roadside infrastructure such as traffic controllers and digital signs, as well as vehicles equipped with vehicle-to-everything technology, commonly known as V2X.

For example, if a vehicle approaches a conflict area — the part of an intersection where different paths cross — at a dangerous speed, EagleEye could keep other approaches on red until it passes. If pedestrians have not finished crossing, the system could hold conflicting traffic for longer. Connected vehicles could also receive warnings about hazards or people their drivers cannot yet see.  

These applications reflect Maselli’s vision for more proactive road safety. Authorities often rely on crash reports to identify dangerous locations, meaning someone may need to be injured before a problem receives attention. Short traffic surveys offer useful information, but they can miss changing road conditions and near misses that occur outside the survey period.

XVision AI aims to give traffic engineers a continuous view of movement, congestion and risk. Near-miss data could help them identify recurring conflicts, adjust signal timing or investigate an intersection before a serious collision occurs. Maselli puts the motivation simply: “One death on the road is too many.”

The company is also addressing the fragmented nature of traffic infrastructure. A single intersection may use road loops, radar, cameras, thermal sensors and analytics software from different suppliers. Bringing these systems takes time and can make upgrades expensive. Maselli estimates that a typical road upgrade in Australia can cost around AUD1.4 million and take 18 months.

EagleEye combines several of these functions inside one unit mounted on an existing pole and connects to the traffic controller. Maselli says installation only take a couple of hours, reducing the need for disruptive civil works. Although EagleEye may cost more than an individual conventional sensor, XVision argues that it can lower the overall cost by replacing several separate systems.

The software behind the hardware may be the company’s strongest commercial advantage. Customers begin with basic traffic analytics and data collection, then activate more applications on the same device. XVision has developed around ten modules, including adaptive signal control and traffic enforcement functions.

Its strongest commercial advantage is, however, the software behind the hardware. Customers begin with basic analytics and data-collection functions, then activate additional applications on the same device. XVision has developed around ten modules, including functions for adaptive signal control and traffic enforcement.

“The important thing to remember is that the hardware is only part of the business,” Maselli says. “Our core business is selling software modules.”

He calls the model “software-defined infrastructure”. Conventional road equipment is usually installed for a fixed purpose and may remain unchanged for years. XVision’s devices can receive new functions through software updates, allowing road authorities to expand their systems without repeatedly replacing roadside hardware.

The wider XVision AI platform includes two other devices. “Scout” is a smaller, solar-compatible sensor designed for quick deployment along roads and corridors. It cannot directly control traffic equipment like EagleEye, but it can collect traffic data and support wireless V2X communication. “Outpost”, on the other hand, places similar technology on a portable trailer, making it suitable for roadworks, temporary traffic studies and work zones. XVision Command brings information from these sites into one platform for monitoring, analysis and reporting.

By June 2026, Maselli said XVision had secured around 190 paid deployments across five Australian states and territories, along with six international pilots. The company has also tested its technology in Vietnam and Thailand.

Nevertheless, expanding into more markets will require more than accurate AI. Traffic enforcement and control systems must meet technical standards that differ between countries and sometimes between states. Some XVision modules, including speed enforcement, red-light enforcement and seatbelt detection, still require regulatory approval before customers can activate them.  

Certification is costly and time-consuming, but it could strengthen XVision’s position once the necessary approvals are in place. Maselli describes the enforcement applications as the company’s “highest-value modules”. He estimates that certification could add AUD25 million in lifetime value across its existing deployments without installing new devices, though this remains the company’s own projection.

“Once we’re certified, we’re no longer competing in that highly competitive segment,” Maselli says. “We move into blue-ocean territory.” In practical terms, approval would allow XVision to compete in a more specialised market with fewer certified rivals.

As with any surveillance technology, privacy presents another challenge. Roadside cameras can make members of the public uncomfortable, even when they are being used for traffic analysis rather than surveillance. To address this, XVision works with customers to display signs explaining why the equipment is present. Its system also blurs video and displays it at low resolution by default, while access to identifiable, high-resolution footage requires additional authorisation.

Maselli views Australia as a proving ground for a much larger market. He sees opportunities in India, Vietnam and Thailand, where governments are investing in traffic technology. The U.S. is another target because of its larger infrastructure budgets and growing demand for connected road systems.

XVision AI still has to navigate government procurement, regulatory approval and competition from established traffic-equipment suppliers. Supporting physical devices across several countries will also be harder than expanding a cloud-based software platform.

Its advantage lies in how the technology was developed. XVision grew from years of solving practical infrastructure problems, including the fragmented system that first convinced Maselli there had to be a simpler approach. If the company succeeds, near misses that once disappeared without a trace could become the evidence road authorities need to prevent the next crash.

Anyone who has opened a carton of milk before its expiry date only to find that it smells off or tastes bitter has experienced the problem firsthand. For dairy processors, the same issue can play out across thousands of litres. Milk may pass routine quality checks, make it through production and still spoil before reaching the end of its expected shelf life.  

CYBERTONGUE^®, a biosensing platform, is designed to give dairy processors an earlier warning.  

Developed by Canberra-based biotech startup PPB Technology, the compact food testing system allows processors to analyse milk on-site and receive results within minutes. Its first commercial application detects protease, an enzyme linked to spoilage, shorter shelf life and production losses across dairy products.  

Dairy companies already test milk for fat, protein, bacteria, antibiotics and other indicators. Samples from individual farms may be sent to central laboratories, while tanker loads are tested again when they reach processing plants. These checks provide useful information, but they do not always show whether milk will remain stable in a UHT carton, meet a supermarket’s shelf-life requirements or produce the expected amount of cheese.  

Even after heat treatment, protease can continue affecting milk. For processors making UHT milk, fresh milk or mature cheese, that creates a costly blind spot.  

“Right now, dairy processors are working blind on this,” PPB Technology founder Stephen Trowell told Ventureport. “It’s becoming more of a problem as customers expect longer shelf lives and higher quality.”  

CYBERTONGUE^® uses engineered biosensing proteins that emit blue and green light. The balance between the two colours changes when the protein reacts with a target substance. The reader measures that shift and calculates how much of the target is present.  

‍

‍

Trowell describes himself as a protein engineer at heart. He spent around three decades at CSIRO, Australia's national science and technology agency, and close to four decades in research before becoming a startup founder. The technology behind CYBERTONGUE^® grew out of years of biological research, including earlier experiments involving wine aromas and explosives detection.

‍

‍

PPB initially developed its technology to measure plasmin, an enzyme derived from cows. After feedback from the dairy industry, the company shifted its focus to bacterial proteases, which were identified as a more pressing spoilage problem. The platform can make some existing tests faster, but PPB chose to focus first on a measurement previously unavailable to processors.  

“Where we’ve chosen to lead is with a measurement nobody else can do at all,” he said. “The blue ocean, if you like.”  

That focused entry point shapes PPB’s wider strategy. CYBERTONGUE^® could eventually test for lactose, allergens, microbial toxins and other targets. Yet launching everything at once would stretch a six-person company too far. Protease gives it a clear route into dairy, where the financial value is easier to demonstrate.  

A processor that detects high protease activity may redirect a batch of milk away from long-life products and use it for yoghurt, curd or another product where the enzyme causes less damage. PPB also says its tests can help trace problems in the milk supply and detect biofilms on processing equipment, allowing factories to take corrective cleaning measures.  

Fresh milk presents another sizeable opportunity, with the global market estimated at around US$81 billion in 2026 and projected to grow at 6.83% annually through 2035. At that scale, even small gains in shelf-life consistency could prevent substantial losses. Supermarkets often require milk to have a minimum amount of shelf life remaining on delivery. Products that fall short may be rejected, discounted or discarded at the processor’s expense. While Australian processors typically aim for a shelf life of 19 to 21 days, they do not always achieve it. CYBERTONGUE^® could help them reach that target more consistently.  

Cheese producers face a related issue. Protease can create bitter flavours in hard cheeses and reduce production yield. Cheesemakers add another enzyme, chymosin, during production, often without knowing how much protease activity is already present in the milk. Better information could help them adjust the process and get more value from the same raw material.  

PPB sells the reader as laboratory equipment, but its long-term business model centres on repeat testing. Customers buy the machine, then purchase a PPB reagent cartridge each time they run an assay. Future tests are intended to work on the same reader, making the system more useful as the catalogue expands.  

CYBERTONGUE^®’s compact design reflects PPB’s focus on practical onsite testing. The device is small enough to fit into two cupped hands and comes with a tablet running PPB’s software.  

‍

‍

At the time of the interview, 13 to 15 systems had been installed with paying customers. PPB had made sales in 11 countries across five continents, with Europe as its largest market.  

Trowell describes PPB as “born global”. While Australia and New Zealand have sophisticated dairy industries, Trowell thinks CYBERTONGUE^® may offer even more value in regions where milk quality is less consistent, transport routes are longer or refrigeration is harder to maintain.  

Trowell also believes PPB has a strong competitive moat, with the combination of manufacturing both the hardware and consumable reagents. Patents form one layer, but he does not consider legal protection sufficient on its own. A competitor would need expertise in protein engineering, reagent production, optical hardware, software and dairy processing.  

Calibration adds another barrier. For each new box of reagents, PPB securely uploads the required calibration data to the customer’s machine. A third party could not simply manufacture a cartridge with the same dimensions and expect it to work correctly.  

“We own the whole stack,” Trowell said, referring to everything from the sensing molecule to the hardware, software and cloud platform.  

That said, he accepts that no technology is impossible to copy. His view is that PPB’s mix of scientific knowledge, integrated system and experience in dairy gives it enough of a head start to build a meaningful position before competitors catch up.  

The technology could eventually reach beyond food. Trowell sees possible uses in environmental testing, veterinary care and medical diagnostics, including the analysis of blood, serum or saliva. PPB has already discussed potential applications with organisations outside the food industry.  

For now, however, moving directly into clinical diagnostics would create more risk than opportunity. Medical products can require years of trials, extensive regulatory work and large development budgets. PPB would also need to build new sales channels and choose between many possible clinical uses without knowing which one would attract the strongest demand.  

For these reasons, food testing offers PPB a clearer path. The regulatory barriers are generally lower, customers have immediate operational problems, and PPB has already built relationships in dairy. Once the platform has proved itself commercially, the company could enter other sectors through partnerships or licensing.  

Trowell has held that view for years. The immediate goal is to build a successful food testing company rather than chase the prestige often attached to medical biotechnology.  

The opportunity may also expand as food production becomes more complex. Dairy factories increasingly process oat, soy and other alternatives on equipment that also handles cow’s milk. This creates cross-contamination and allergen risks, offering another potential use for CYBERTONGUE^®.  

PPB still has plenty to prove. It must secure repeat reagent orders, convert early trials into wider deployments and expand its test catalogue without compromising accuracy. Scaling an integrated biotech hardware business across international markets will not be simple.  

Still, its approach offers a useful lesson for other deep-tech startups. Instead of beginning with every possible application, PPB picked a specific problem that costs customers money today. It built a product around that pain point, then used the same technology as the foundation for a broader platform.  

As Trowell put it, customers “need it fast, and they need it right.” CYBERTONGUE^® is betting that food testing can deliver both.  

Anyone who has snorkeled over a coral reef has probably admired its vibrant colours and the marine life swimming between its branches. What is less visible is how easily these ecosystems can be damaged. Rising ocean temperatures and pollution continue to put reefs under pressure, while coastal development creates another threat. Construction can stir up sediment that settles on corals, blocking the sunlight they need and making it harder for damaged reefs to recover.  

Protecting marine ecosystems is becoming a shared responsibility. Alongside scientists, governments and conservation groups, businesses whose operations affect the ocean are increasingly expected to understand, manage and reduce their environmental impact. They must also show regulators, banks and investors that they are meeting their environmental commitments.  

Archireef helps businesses meet those demands. Founded by Vriko Yu, the Hong Kong-rooted B2B startup specializes in marine ecosystem restoration and ocean nature risk management. Drawing on years of research at the University of Hong Kong, Yu built the company to help organisations understand, measure and reduce their impact on marine ecosystems.  

Yu describes Archireef as an “end-to-end ocean nature risk management company”.

“What we offer them is a credible way to meet those obligations and stand behind the results,” she says.

One of the company’s main tools for restoring damaged marine ecosystems is its flagship 3D-printed terracotta reef tile. The tiles provide a stable foundation where young corals can attach, grow and survive over time.  

Archireef chose terracotta instead of concrete and other conventional materials because it is pH-neutral and does not release harmful chemicals into the water. Its surface also gives corals a suitable place to attach naturally.

Before placing a single tile in the ocean, the team studies the proposed restoration site to determine whether recovery is possible. It assesses the water quality, biodiversity and general health of the ecosystem. As Yu explains, restoration should begin only where nature has a realistic chance to recover.

“Restoration isn’t about replacing nature or engineering a reef into existence. It is about removing the barriers and giving a degraded system the conditions it needs to begin recovering on its own”, she says.

Once the team selects a site, divers carefully place the tiles on the seabed. Their curves, ridges and crevices imitate the structure of a natural coral reef, creating sheltered spaces for marine species. The design also helps limit sediment build-up, which can block sunlight and smother young corals.

‍

‍

Archireef adjusts the shape of its reef tiles for each project.

“There’s no universal reef, so there’s no universal tile”, Yu says.

Instead of producing one standard model, the company adapts each design to the coral species, water conditions and surrounding marine environment. A coral restoration project in Hong Kong may therefore require a different tile from one in Saudi Arabia, even when the same manufacturing process is used.

“The biggest adaptation from site to site is usually the match between design, species and environment, not the manufacturing itself ”, Yu explains.

‍

‍

Although the reef tile is Archireef’s most recognizable innovation, Yu sees it as one part of a much broader service.

“The reef tile gets the attention because it’s tangible,” she says. “But the real product is the proof — defensible, audit-ready evidence of nature recovery that an organization can put in a disclosure and stand behind under scrutiny”.

Producing that evidence requires years of biodiversity monitoring. Long after the reef tiles have been installed, Archireef continues to track coral survival and the health of the surrounding marine ecosystem. Its monitoring methods include ecological surveys, AI-assisted species recognition, photogrammetry and environmental DNA testing.

‍

‍

Photogrammetry uses photographs to produce detailed 3D models of the reef. These models allow the team to measure changes in coral growth and reef structure over time.

Environmental DNA, commonly known as eDNA, offers another way to monitor marine biodiversity. Traditional ecological surveys often depend on divers recording the species they can see. eDNA testing instead examines traces of genetic material left in the water by marine organisms, including skin cells, mucus and waste.

By analyzing these traces, Archireef can detect species that may be hidden, difficult to identify or absent during a dive. This gives the company a more complete picture of the ecosystem and how it is changing.

For Archireef, the success of a coral reef restoration project is determined by what happens several years after installation.

“The indicators that matter aren’t the ones that look good on deployment day”, Yu says. “They’re the ones that show, two, three and five years on, whether a genuine ecosystem is establishing and sustaining itself”.  

At the company’s first restoration site in Hoi Ha Wan, Hong Kong, around 90% of the transplanted corals remained alive four years after deployment. The long-term survival rate suggests that the ecosystem is recovering rather than simply appearing healthy during the early stages of the project.

Results like these have helped Archireef move beyond academic research and secure collaborations across the public and private sectors. In Hong Kong, the company has worked with government departments and public organisations responsible for coastal infrastructure and marine conservation.

Most B2B sales teams spend their days inside software. They track leads in a customer relationship management (CRM) tool, send emails, build outbound lists and measure every digital touchpoint. Still, many important business relationships move forward in person. A dinner after a conference, a small founder roundtable or a private customer event can do what dozens of cold emails cannot: build trust quickly.  

That is the gap Marvelous wants to fill. The bootstrapped San Francisco-based AI startup is building what founder and CEO Merve Isler calls “Salesforce, but for real life”. Her idea is simple: if companies already use software to manage digital sales, they should also have software to manage the real-world moments that help close deals.  

Marvelous brings data, automation and targeting into in-person go-to-market (GTM) work. It helps B2B sales, revenue, marketing and growth teams plan in-person events such as curated dinners, launch events, mixers, happy hours and conference side events. These formats are already familiar to companies selling to enterprise buyers. The problem is that they are often managed through spreadsheets, venue calls, scattered guest lists and manual follow-ups.  

That creates two problems. First, the process is slow. Teams can spend weeks figuring out who to invite, where to host and how to manage responses. Second, the return is hard to measure. Companies spend heavily on conferences, dinners and private events, yet they often do not have a clear way to know whether an event will influence pipeline before they commit the budget.  

Marvelous wants to pull those pieces into one AI-powered event management platform.  

“Our main target customer is revenue and sales teams,” Isler said. “Because about 40% of deals close in person, but there's no infrastructure built for that—so that's what we do.”  

For Isler, the idea comes from experience. Before starting Marvelous, she worked at Google from Istanbul, managing developer product launches and go-to-market programs across Turkey, Central Asia and the Caucasus. During that time, she helped build more than 100 communities around the world and coordinated thousands of events a year across eleven time zones. The work was fast, local and very manual.

That experience shaped the foundation for Marvelous. Isler saw that building products had become easier, while distribution remained difficult. A startup can now ship software faster than ever. Reaching the right people, in the right setting, is still a different challenge.  

At Google, Isler had to understand how people gathered, communicated and built trust in different markets. A product launch might require developer meetups in dozens of cities, but the right approach could change from country to country. In Kazakhstan, she said, Telegram worked better than WhatsApp. In Afghanistan, Facebook mattered more. Each market had its own habits, and growth depended on understanding those details. Those details and adapt quickly.  

Marvelous grew out of that playbook. Isler saw that offline distribution had patterns, even when it looked chaotic from the outside. The right guest list, the right room, the right timing and the right follow-up could change the outcome of a sales conversation. Most of that knowledge, however, lived in people’s heads. Marvelous is her attempt to turn it into software.  

Inside the platform, a company can start by choosing the kind of satellite event it wants to host: a launch event, a mixer, a happy hour, a brunch, a lunch or a dinner. From there, users can connect their CRM. When Salesforce is connected, Marvelous analyzes contacts and creates relationship scores based on signals such as buying intent, past event activity and how warm a relationship appears to be. If a team also uses an event platform such as Luma or Eventbrite, Marvelous can bring that event data into the picture as well.  

‍

‍

‍

The product is built around Maven, Marvelous’ AI assistant. Maven can help plan an event through Slack, iMessage or the Marvelous platform itself. For instance, a user might ask Maven to plan an executive dinner for 18 people within a set budget. From there, Maven can find warm contacts, estimate who is likely to attend, build a guest list, recommend venues, send invitations and run follow-up sequences across email, LinkedIn and SMS. The goal is to let sales teams focus on conversations and closing deals instead of worrying about logistics. Put simply, Marvelous helps a company find the right guests, secure the right venue and understand which event format is likely to work. Instead of guessing whether a US$10,000 dinner will pay off, the company gets a clearer view of the potential return.

‍

‍

AI Insiders, Marvelous’ invite-only network of verified AI and enterprise leaders, is also part of the company’s go-to-market strategy. Rather than asking every customer to build an event from scratch, Marvelous can group several companies that want to reach the same audience into one curated event. Such a setting creates revenue, product feedback and fresh data about what works in different markets. The network spans verticals such as cybersecurity, fintech, robotics, healthcare AI and enterprise SaaS. It now includes members from more than 475 companies, with C-level executives making up nearly half of the community. So far, AI Insiders has facilitated more than 2,000 introductions and contributed to over US$560 million in deals. Isler describes it as one of Marvelous’ strongest go-to-market channels.  

The first AI Insiders event took place at AWS GenAI Loft in San Francisco in March 2025. It brought together 150 curated guests, including AI founders, tier-one investors, enterprise executives and researchers for an evening focused on high-impact conversations and collaboration.  

The timing may work in Marvelous’ favor: Digital outreach is getting louder, and AI will make it easier for companies to send more emails, messages and automated pitches. That may make high-trust in-person conversations more valuable, especially in enterprise sales where relationships take time. At the same time, event budgets need proof. Revenue leaders want to know whether a dinner, roadshow or customer event is worth the cost.  

Marvelous is betting that offline sales will become a measurable category for software. If the company can prove that its relationship scores, event intelligence and AI agent help teams create a stronger pipeline, it could become an important tool for B2B go-to-market teams. The core message of Marvelous is easy to understand: the deals may happen in the room, but data can still power the room.