At Wolfram, we are passionate about delivering extraordinary results

Radio Access Networks (RANs) form the essential bridge between mobile devices and a network’s core infrastructure. For 3G and 4G networks, mobile operators have traditionally relied on proprietary RAN systems—typically sourced from a single vendor—which can limit interoperability, raise long-term costs and slow adaptation to new technical demands.

In the shift to 5G, Wolfram has shown how AI-driven network optimisation and a deployable full-stack environment extend what Open RAN can deliver, giving developers direct access to built-in machine learning and advanced visualisation inside the network.

That vision of a more open and adaptable network underpins Wolfram’s work in the Cambridgeshire Open RAN Ecosystem (CORE) project. The initiative—one of 19 regional trials funded through the UK government’s Open Networks Ecosystem program—served as a live, multi-vendor testbed.

Wolfram’s role centred on building and deploying a predictive optimisation rApp inside this multi-vendor network, demonstrating how machine learning can guide real-time decisions and proving that the full Wolfram stack can operate natively in an Open RAN environment.

An Open Way Forward

As 5G networks evolve to support increasingly complex demands—from real-time media streaming to dense device connectivity—open, interoperable infrastructure offers a path toward more adaptive and cost-effective deployments. Countries like the United States and Japan have already begun scaling Open RAN technologies in live environments, while adoption in Europe is still in the early stages.

Wolfram collaborated with eight partners ranging from infrastructure providers to academic researchers to deploy a testbed Open RAN 5G network in Cambridgeshire. Unlike prior lab demonstrations, the CORE network operated across public venues and supported hardware and software from multiple vendors, demonstrating the viability of a modular, interoperable network at scale. In field testing, Wolfram’s predictive optimisation rApp helped the network sustain download speeds of 700 Mbps with latency under 10 milliseconds—performance levels proven during a live augmented-reality concert demo.

Wolfram’s role in this project centred on building an rApp within the RAN Intelligent Controller, showing how AI-driven insight can support real-time policy decisions without sacrificing explainability or control. Just as significant, the project proved that the full Wolfram technology stack can run natively inside an Open RAN environment, giving developers immediate access to advanced computation previously available only outside the network. As UK operators explore new architectures and supply-chain diversification, the CORE project stands as a concrete example of what’s possible—and what’s ready to scale.

Wolfram’s Network Optimisation rApp

In the Open RAN architecture, rApps are standardised software applications that run in the non-real-time RAN Intelligent Controller (Non-RT RIC). Their role is to analyse historical and near-real-time network data to support intelligent traffic management, anomaly detection and policy optimisation—helping operators adapt to changing network conditions without manual intervention.

rApps are a core component of the O-RAN specification, and Wolfram developed a native Wolfram Language implementation for the CORE project. The result was a customised rApp that provided predictive analytics and explainable decision support within a multi-vendor environment. It also established a foundational Open RAN SDK to drive future data and AI-powered app development.

Wolfram’s Network Optimisation rApp is a full-stack implementation designed to support intelligent decision-making in Open RAN environments. Built using Wolfram Language, the system combines a back-end machine learning engine that analyses network data and predicts performance patterns with a front-end dashboard that visualises key metrics and delivers actionable recommendations to engineers.

Developed as part of the CORE project, the rApp was successfully deployed alongside partner hardware in a multi-vendor testbed. This deployment also confirmed that the entire Wolfram technology stack could operate natively within the RAN environment, giving developers direct access to advanced computation and machine learning without external integration.

The back end of Wolfram’s rApp processes live network performance data to support intelligent optimisation. It ingests key performance indicators (KPIs) such as throughput, latency, signal strength and packet loss—core metrics used to assess the health and efficiency of a 5G network. Using these inputs, the machine learning model identifies patterns in traffic behavior, detects anomalies and generates recommendations aimed at improving network performance. These outputs are designed to help operators proactively manage traffic flow and resource allocation across the network. While retraining frequency depends on deployment parameters, the model is intended to operate continuously in tandem with incoming telemetry data.

The front-end dashboard of Wolfram’s rApp provides a visual interface for monitoring network performance in real time. Engineers can view KPIs such as throughput and signal strength using standard visualisations, with options to compare multiple data streams or focus on specific time windows. Users have control over which metrics are displayed and can adjust the timeframe to analyse short-term fluctuations or longer-term trends.

One view, for example, displays throughput across two adjacent cells, offering a side-by-side comparison that helps operators quickly identify load imbalances or performance deviations. This interface supports decision-making by making underlying patterns immediately visible and aligning visualisations with actionable insights generated by the back end.

The rApp’s recommendation system translates back-end analysis into actionable suggestions for network engineers. When the model identifies a configuration that could improve performance—such as activating additional cells to handle rising traffic—it surfaces a recommendation through the dashboard interface. These suggestions are presented alongside relevant context and supporting metrics but are not applied automatically; engineers remain responsible for reviewing and approving any changes. Here is an example recommendation detailing a traffic-steering policy for a specific cell based on predicted demand.

The Promise of Open 5G: Higher Speeds, Exciting Applications

Tests conducted during the CORE project trial in Cambridgeshire demonstrated that the open 5G network consistently achieved download speeds exceeding 700 Mbps with sub-10-millisecond latency and 100% uptime during live operation. These results were obtained using commercially available devices under typical deployment conditions, not lab-simulated scenarios.

Performance was well above the ITU’s recommended 5G benchmark of 100 Mbps for user-experienced throughput and outperformed trial speeds reported in similar UK Open RAN pilots, such as Three UK’s 520 Mbps deployment in Glasgow—confirming that Open RAN configurations can deliver production-grade connectivity in high-demand settings.

To evaluate network performance under immersive, high-bandwidth conditions, the CORE consortium organised a public augmented-reality concert trial. A live performance at the Cambridge Corn Exchange was streamed in real time to a second location, where participants wearing Meta Quest and Apple Vision Pro headsets viewed the show as spatial 3D video with synchronised audio.

The feed—captured using custom 8K 3D cameras and streamed over the open 5G network—was delivered with less than 1.5 seconds of latency and maintained flawless audio-visual sync across multiple headsets. This setup provided a realistic stress test of the network’s capacity to handle sustained, low-latency, multi-device loads in a public setting.

The successful execution of the augmented-reality concert trial underscores the potential of Open RAN to support next-generation applications that demand both high throughput and ultra-low latency. It also validated the ability of a multi-vendor, standards-compliant architecture to deliver consistent performance under real-world load conditions.

While this trial was limited in geographic scope, the results provide a working proof of concept for scalable, interoperable network design. By adding software-defined intelligence through Wolfram’s rApp, the project showed that intelligent optimisation and real-time responsiveness can coexist—a model for broader adoption of adaptive 5G deployments across the UK.

Ready to Build Smarter Networks?

The CORE Project proved two things: that Wolfram’s full technology stack can run natively inside an Open RAN environment—powering rApps, dashboards and machine learning–driven optimisation—and that Wolfram can deliver those results while working closely with industry partners and government-led initiatives. Together, these capabilities position Wolfram Consulting as a partner ready to help operators and vendors turn network data into real-time intelligence and adaptive 5G solutions.

Contact Wolfram Consulting to put the full Wolfram tech stack to work inside Open RAN—delivering predictive optimisation and actionable insights out of the box.

Capgemini Engineering and Wolfram Research are developing a co-scientist framework: a tool designed to support engineers working on complex physical systems. Drawing on shared strengths in symbolic computation, generative AI and systems engineering, the co-scientist helps translate engineering intent into executable, verifiable computations.

This project is part of Capgemini’s broader Augmented Engineering strategy, an initiative that applies hybrid AI techniques to real-world engineering challenges. By combining Wolfram’s expertise in symbolic computation with generative AI, the co-scientist helps teams engage with complex problems earlier in the design process—making it easier to refine assumptions and address critical risks before they compound.

Natural Language In, Symbolic Logic Out

The co-scientist is designed to bridge natural language input and computational output. By combining large language models with Wolfram’s symbolic computation and curated knowledgebase, it allows engineers to ask domain-specific questions in plain language and receive results in the form of executable code, equations or dynamic models that can be verified and reused.

Using the co-scientist feels less like querying a search engine and more like working with a technically fluent collaborator. Engineers can describe the problem in their own words—“simulate thermal behavior under load” or “optimize actuator response time”—and receive results they can validate immediately. Behind the scenes, the co-scientist generates Wolfram Language code and combines existing models, Wolfram Language code and computable data to return outputs that are not just plausible but logically structured and derived from verifiable computation.

Building Trust with Hybrid AI

This collaboration puts hybrid AI to practical use by combining the flexibility of language models with the structure of symbolic reasoning and computational modeling. Instead of generating surface-level responses, the system can explain its logic and return results that hold up in real-world settings like aerospace or industrial automation.

In regulated or high-risk environments, outputs must be traceable and reproducible. Hybrid AI systems built on symbolic foundations allow teams to audit every step: how an equation was derived, what assumptions were embedded and how the result fits within engineering constraints. That’s not just helpful—it’s required in science and engineering.

From Acceleration to Transformation

The co-scientist doesn’t just accelerate existing workflows—it shifts how engineers and researchers approach complex challenges. Instead of translating a question into technical requirements and then into code, users can work at the level of intent, refining the problem itself as they explore possible solutions. That creates space for earlier insights, faster course corrections and a more iterative, computationally grounded design process.

As this collaboration continues, Wolfram’s computational framework keeps the co-scientist anchored in formal logic and verifiable output—qualities that matter in domains where failure isn’t an option. From engineering design to sustainability analysis, the co-scientist is already showing how generative AI can shift from surface-level response to real computational utility.

Cancer researchers aren’t short on data, but they are often overwhelmed by it. The Cancer Genome Atlas (TCGA) contains more than 2.5 petabytes of genomic, clinical and imaging data across 33 cancer types. For those in the medical and research communities—especially without formal computational training—making practical use of that data remains a significant challenge.

As a result, a valuable resource remains underutilized.

Dr. Jane Shen-Gunther, a gynecologic oncologist and researcher with expertise in computational genomics, encountered these limitations firsthand. Drawing on her deep expertise, she recognized the pressing need for a more accessible, unified method to access and analyze genomic and imaging data from multiple sources.

“The TCGA, The Cancer Imaging Archive (TCIA) and the Genomic Data Commons (GDC) are essentially three goldmines of cancer data,” she said. “However, the data have been underutilized by researchers due to data access barriers. I wanted to break down this barrier.”

Shen-Gunther partnered with Wolfram Consulting Group to design the TCGADataTool, a Wolfram Language–based interface that simplifies access to TCGA cancer datasets.

Designing a Research-Ready Tool with Wolfram Consulting

Drawing on prior experience with the consulting team, Shen-Gunther worked with Wolfram developers to design a custom paclet to streamline data access and clinical research workflows while remaining usable for individuals without extensive programming experience.

Shen-Gunther explained, “Mathematica can handle almost all file types, so this was vital for the success of the project. It also has advanced machine learning functions (predictive modeling), statistical functions that can analyze, visualize, animate and model the data.”

Interface and Data Retrieval — The guided interface (TCGADataToolUserInterface) allows users to select datasets, review available properties and launch key functions without writing code. Built-in routines support batch retrieval of genomic data from the GDC and imaging data from TCIA, including scans and histological slides associated with TCGA studies.

Data Preparation and Modeling — Processing functions like cleanRawData and pullDataSlice prepare structured inputs for analysis by standardizing formats and isolating relevant variables. Modeling tools enable users to identify potential predictors, visualize candidate features, generate design matrices and build models entirely within the Wolfram environment.

Visualization Tools — The paclet includes support for swimmer plots, overall survival plots and progression-free survival plots, helping researchers visualize clinical outcomes and stratify cases by disease progression or treatment response.

By combining domain-specific requirements with Wolfram Language’s built-in support for data processing and analysis, the paclet makes it possible for researchers without programming experience to work with genomic and imaging datasets that would otherwise require advanced technical skills. This shift enables researchers without programming experience to work directly with TCGA data using computational methods previously out of reach.

Moving from Complexity to Capability

The TCGADataTool was developed in response to specific challenges Shen-Gunther encountered while working with genomic and imaging data. She emphasized the importance of designing a tool that could support clinical research workflows without requiring a background in programming—and saw the result as both accessible and technically robust.

By designing a tool that supports clinical research without requiring fluency in code, Wolfram Consulting Group delivered a solution tailored to the needs of domain experts—extending the reach of high-throughput data into contexts where traditional software workflows often fall short.

“The [Wolfram team] carefully and thoughtfully developed technical solutions at every step and created a beautiful, easily accessible product,” said Shen-Gunther. “Their expertise in data science and user-interface development was essential to the success of the paclet.”

Rather than offering a generalized platform, Wolfram Consulting delivered a focused, researcher-specific application that addresses both technical complexity and day-to-day usability. That model—identifying key obstacles, understanding the research workflow and delivering a targeted solution—can be extended to other biomedical contexts where access to large datasets is essential but often limited by tool complexity.

If your team is facing similar data access or analysis challenges, contact Wolfram Consulting Group for a solution tailored to your research environment.

AGM, a globally operating digital marketing agency, develops advertising strategies and executes online marketing campaigns for its customers from a broad range of sectors. Their challenge was to determine the best possible allocation of marketing funds among multiple online channels, optimizing the overall effectiveness and return of investment of its marketing campaigns.

Offshore wind is one of the most important sources of renewable energy and a key area of interest for one of Wolfram Consulting Group’s clients. To get a complete understanding of the multitude of factors that contribute to the technical and financial performance of a wind farm, our client’s challenge was to design and develop a complete software package for modeling offshore wind operations.

From the beginning, the founders of the biotechnology startup Emerald Therapeutics wanted to develop an ideal research platform that would allow for lab and process automation during experiments as well as easy communication of their findings.

Mathematica provides all the tools the Emerald founders needed to carry out their vision, while Wolfram Language powers Emerald Cloud Laboratory, a web-based life science lab.

A major US-based pharmacy chain needed to reliably automate the process of converting doctors’ prescriptions to standardized printed labels that are attached to customers’ prescription medicines. The doctors’ notes, written in expert English abundant with idioms and conventions used in medical practice, amounted to more than 100 million unique strings of text to be processed as a data stream with minimal latency at tens of transactions per second.

When you need to boost performance, reduce operating costs or investigate design space, model-based design is a powerful tool. With model-based design, virtual models are at the center of the development process and help you shorten development cycles and substantially reduce overall development costs.

Langham Hall, a global professional services business in the fund management industry with assets of $130 billion, partnered with Wolfram Consulting Group to create a custom platform for its clients. The goal was to build a system with unparalleled personalization, accuracy and reliability without compromising speed or efficiency. This required an innovative approach from experts in data science and big data, which Wolfram, a leader in computational intelligence across industries, could provide.