Rebuilding the Roaming Platform a Billion People Rely On — Without Taking It Offline

A global telecom analytics provider needed to modernize the 20-year-old system that steers roaming traffic for mobile operators worldwide. 3Pillar is doing it with HelixAI — while the platform stays live, and without disrupting the operators that depend on it.

Every time a phone leaves its home network, a decision has to be made in real time: which partner network should carry that call or data session, at the best balance of cost and quality, without dropping the connection or letting fraud slip through. For more than a billion people each year, that decision runs through a single platform — built and operated by 3Pillar’s client, a global telecom software provider, and trusted by mobile operators around the world.

That platform had carried the responsibility for more than twenty years. Close to 400 engineers shaped and extended it across its lifetime, and over time it became what many long-lived, mission-critical systems become: indispensable, deeply complex, and layered with two decades of accumulated logic, with the rules for some 200 mobile operators woven into a single codebase. Much of its original architecture had never been fully documented, and a great deal of how it worked lived only in the code itself. Yet the future the business wanted to build — continued growth in 5G and IoT, and new AI-native capabilities — depended entirely on first making that system safe to change.

This was modernization under demanding conditions. Roaming traffic never pauses, so the rebuilt platform had to match or exceed the original on performance, availability, resilience, and security — with no interruption to live service, and no compromise to the standards and operator agreements that govern how calls and data move between networks. The client’s own engineers, meanwhile, were fully committed to the product roadmap their customers were waiting on; they could guide the work, but they could not pause to do it themselves.

The modernization had to move at the pace the business demanded, around a team that never stopped, and without ever taking the system offline.

3Pillar began the way every sound modernization should: by replacing assumptions with understanding. Before proposing a plan, the team put ATLAS to work — the continuously updated intelligence layer of HelixAI that reads source code in a read-only environment and maps a system’s structure, its dependencies, and the business logic behind them.

In just three weeks, ATLAS mapped a codebase that had taken two decades and hundreds of engineers to build, and the team turned that understanding into a 90-page modernization plan — one with a clarity and precision that interview-driven assessments rarely reach, produced without ever touching a live system. Along the way, ATLAS revealed that as much as 40% of the legacy code was dead or unreachable, work that could be set aside before modernization even began.

Discovery delivered a second, larger advantage. Several capabilities the team had expected to rebuild from scratch — including parts of the signaling layer — already existed elsewhere in the client’s own systems and could be adapted instead. That single insight cut the estimated effort on those components by 50 to 60%, and it set the tone for the engagement: understand the system deeply, then build only what genuinely needs building.

With the system mapped, 3Pillar brought in Helix Pods — HelixAI’s AI-native delivery teams — to carry the work forward on two fronts at once. One pod modernizes the core engine, the part that makes the real-time steering decisions, porting it from legacy C++ to modern Java while preserving its behavior exactly. The other strengthens everything beneath it: the build systems, security, configuration management, and the data and caching layers that keep the platform fast and reliable.

Running both tracks in parallel is what turns a multi-year undertaking into a focused, roughly twenty-week program. And because 3Pillar leads the modernization directly, the client’s own engineers stay free to drive the roadmap their customers are waiting for — with a clear plan to hand the modernized system, fully documented, back to their team at the finish.

Moving quickly across a system this sensitive demands an equal measure of control, and that is the role of AIRE, HelixAI’s AI Responsible Engineering framework. AIRE keeps every use of AI across the software development lifecycle explicit, constrained, and inspectable. Before any change reaches the signaling and steering logic operators depend on, the team captures exactly how the system behaves today in characterization and regression tests, establishing a baseline to prove every change against. And because each change ships behind a feature flag, any step can be reversed in moments if it needs to be.

That discipline is precisely what makes it possible to modernize a platform carrying live roaming traffic without ever leaving the business exposed.

The momentum behind this program comes not from any single tool, but from the way HelixAI’s parts reinforce one another. ATLAS doesn’t go quiet once discovery ends; it keeps running, maintaining a living map of the system and surfacing dead code and shifting dependencies as the work moves forward. The Helix Pods build against that current intelligence rather than against guesswork, so engineering effort lands exactly where it counts. And AIRE governs every change against the behavioral baseline, holding firm on the standards and agreements the business runs on.

Discovery informs delivery, delivery is governed by design, and that governance gives the team the confidence to keep moving at pace. The result is a modernization that stays visible and controllable from end to end — a guided program, rather than a sequence of high-stakes changes made on incomplete information.

This is a telecom story, but the challenge at its heart is not unique to telecom. Almost every established enterprise carries a system like this one — too critical to retire, too tangled to change safely, quietly holding back the work around it. The easy choice is to leave it alone, and the cost of that choice only compounds with time.

What this engagement shows is that a system like this doesn’t have to stay untouchable. With deep discovery up front, a disciplined way of delivering the work, and governance built in from the very first change, even a platform two decades in the making can be modernized safely — and at the pace the business actually needs.