Rail · Predictive Maintenance

Theia

"See failure before it becomes disruption."

Rail operators had fleet and infrastructure condition data scattered across separate inspection and maintenance systems, making it hard to catch a developing failure before it became an outage.

THScreenshot coming soon

How to read this case study: what I owned, what shipped and the verified outcome (My Role, What Shipped and Impact below) are factual. The surrounding strategy, vision, alternatives and metrics framework are interview-ready framing built on those facts, how I'd talk through the product thinking, not a claim that every metric or GTM motion was formally run at the time.

The problem

Challenges & baseline.

Challenges

Maintenance insight lived across disconnected inspection and monitoring systems, so operators were reacting to failures rather than catching them early, exactly the gap Theia was built to close. The problem was never an absence of information, operators already had inspection and condition data, it was that the information was fragmented, making a single picture of asset condition hard to build.

Baseline

Asset condition visibility before Theia: fragmented across separate systems, reviewed manually.

The insight

What the customer actually needed.

Insight

The fundamental need wasn't "give me more asset data", it was "tell me which assets need attention, why, how urgent it is, and what I should do before it affects the railway". That distinction matters: the product has to compete on decision quality and time-to-action, not on how many dashboards it adds.

Job to be done

When an asset starts deteriorating, help me identify and understand the issue early enough that I can intervene before it becomes an operational failure.

The strategy

Vision, mission & the bets we made.

Vision

A railway where developing asset failures are identified early enough to prevent disruption.

Mission

Turn fragmented rail inspection and monitoring data into trusted, actionable maintenance intelligence.

Objectives

Replace manual visual-inspection review with automated, machine-learning-assisted image analysis; give operators a contextual, 3D view of asset condition instead of a flat report; keep the platform modular so it scales from a single fleet to a national network.

Goals

Ship real-time fleet and infrastructure condition dashboards; ship automated image analysis for visual inspections; ship 3D digital twin visualisation for asset diagnostics; keep the architecture API-driven so it fits into operators' existing systems.

Strategic bets

  • Consolidate before adding more interfaces, engineers don't need another isolated monitoring system, they need Theia to become the analytical layer connecting the sources they already had
  • Move from monitoring to prediction, reporting what already happened creates limited differentiated value, the higher-value problem is helping operators anticipate what's likely to happen next
  • Integrate into existing maintenance ecosystems via an API-driven architecture, enterprise rail operators rarely replace their entire operating environment for one product
  • Build trust into the AI itself, not just the alerts, since false positives create alert fatigue and false negatives have real operational consequences, precision, recall and confidence became product decisions, not just technical ones

Alternatives considered

Three directions were on the table. Building more individual monitoring applications was rejected, that leaves the fragmentation problem untouched. Building dashboards only was rejected too, it consolidates information but doesn't meaningfully improve the decision itself. The direction that fit the actual job was an intelligence layer, consolidate the data sources first, then progressively add analytical and predictive intelligence on top, rather than trying to ship prediction before the underlying data was even in one place.

My role

What I owned.

Product Owner at Camlin, part of the product team shaping Theia's roadmap and prioritisation within Camlin Rail's portfolio.

Outcomes

What shipped.

Theia gives rail operators a consolidated, cloud-based view of fleet and infrastructure condition, with automated image analysis and 3D digital twin visualisation replacing manual inspection review, scaling from smaller operators up to national networks.

Execution

How it got built.

01

Mapped how condition and inspection data already flowed across operators' separate systems before designing the consolidation layer

02

Shipped real-time fleet and infrastructure condition dashboards as the first consolidated view

03

Added automated, machine-learning-assisted image analysis to replace manual visual-inspection review

04

Layered in 3D digital twin visualisation so asset context, not just a flat reading, was available for diagnostics

05

Kept the architecture API-driven throughout so it could sit inside an operator's existing maintenance workflow rather than requiring a replacement

Go-to-market

How this reaches customers.

This is fundamentally an enterprise land-and-expand motion, not a mass-market launch. Land: pick one operator, one fleet, one inspection problem, one high-value asset category, and demonstrate a measurable improvement against the existing inspection process. Prove: build the business case explicitly, inspection data leads to earlier detection, earlier detection leads to earlier intervention, earlier intervention leads to avoided disruption and financial value. Integrate: connect Theia into the customer's existing asset or maintenance systems so it becomes part of the workflow, not a parallel one. Expand: once credibility is established on one asset category, grow to multiple assets, then the fleet, then infrastructure, then the wider network.

Metrics

North star metric.

Actionable defects caught earlyRecommended North Star metric: percentage of actionable developing defects identified before operational impact. No independently verified percentage is published here, this names the metric Theia should be measured by, not a claimed historical result, see Metrics above for the full framework.

Supporting metrics

  • Recommended North Star going forward: percentage of actionable developing defects identified before operational impact, rather than feature or dashboard count
  • Detection lead time and defect precision/recall as the core decision-quality metrics
  • Time from detection to a maintenance decision, since a fast, trustworthy alert with no clear next action doesn't reduce unplanned maintenance
  • Engineer/platform adoption among the teams the alerts are actually meant for

Guardrails

  • False-negative rate, missing a real defect is the costliest failure mode
  • False-positive rate, since alert fatigue quietly kills adoption
  • System availability and data latency
  • Model confidence and integration failures with operators' existing systems
Result

The verified bottom line.

Positions Theia as the analytical engine behind Camlin Rail's inspection products, shifting operators from reactive, manual reviews toward a predictive, consolidated view of asset condition across the network.

Learning

What I'd do differently.

Enterprise infrastructure software earns trust through consolidation, not more dashboards. The operators Theia serves didn't need another system to check, they needed the existing ones connected into one dependable view, which is why keeping the architecture API-driven and modular mattered as much as any single feature. The other lesson: sophisticated technology doesn't automatically create a sophisticated product, in predictive maintenance value only appears once complex data is translated into a decision that fits the engineer's actual workflow.

More proof, more products.

See every product shipped and business built.