.svg){kind=small}
.svg)
.svg)
Decision Intelligence for Global Gas Infrastructure An AI-powered lakehouse architecture that synchronizes real-time telemetry with engineering specifications to optimize network performance.
The Result: Detected and quantified volumetric measurement inaccuracies that revealed €1 million per year in unmeasured distributed gas. The system shifted the organization from reactive reporting to proactive engineering intelligence, directly impacting capital allocation and infrastructure right-sizing.
AI for Gas Infrastructure Evaluation is an AI-powered decision intelligence system designed to help gas network operators understand, optimize, and financially quantify the performance of their infrastructure.The platform integrates two critical data domains within a scalable lakehouse architecture:Real-time telemetry from pumps, compressors, and field equipmentEngineering datasheets and operational curves, processed, vectorized, and stored for retrieval-augmented reasoning (RAG)Rather than functioning as a traditional monitoring dashboard, the system enables engineering-grade reasoning. Stakeholders interact with an AI chatbot that queries both live operational data and equipment design specifications simultaneously. This allows the system to evaluate how assets are operating relative to how they were designed to operate.By combining telemetry with technical specifications, the AI identifies systemic inefficiencies, equipment mismatches, and measurement inconsistencies across the network. For example, through integrated analysis of flow measurements and operating envelopes, we detected and quantified volumetric measurement inaccuracies across the entire infrastructure.This revealed approximately €1 million per year in unmeasured distributed gas.The key value lies in transforming fragmented engineering data into actionable insight:- Quantification of hidden financial leakage- Improved measurement reliability and compliance- Infrastructure right-sizing and operational alignment- Data-driven capital allocation and performance optimizationThe system is scalable, maintainable, and extensible across additional assets and infrastructure layers. It shifts gas infrastructure management from reactive reporting toward proactive, AI-assisted engineering intelligence—where operational decisions are grounded in both real-time behavior and technical design reality.
This project solves a critical visibility, efficiency, and financial control problem for gas infrastructure operators.Gas networks generate vast amounts of telemetry from pumps, compressors, and flow meters. Separately, engineering specifications define how those assets are designed to operate. In most organizations, these data domains remain siloed. As a result, operators can monitor performance but cannot systematically reason about whether assets are operating within their intended design envelopes.This fragmentation creates three core problems.First, hidden financial leakage. Measurement inaccuracies, calibration drift, and assets operating outside optimal ranges lead to discrepancies between distributed and recorded gas volumes. These losses often remain undetected because there is no integrated system capable of validating operational behavior against technical specifications. By combining telemetry and equipment data, the project enables detection and quantification of systemic volumetric measurement errors—revealing approximately €1 million per year in unmeasured distributed gas.Second, reactive decision-making. Traditional dashboards report metrics but do not provide contextual reasoning. Engineers must manually correlate operating curves, load profiles, and measurement behavior to identify inefficiencies. This is time-consuming, inconsistent, and dependent on individual expertise. The AI system provides structured, engineering-grade reasoning that scales beyond individual knowledge.Third, capital inefficiency. Oversized or poorly matched infrastructure can introduce energy losses, inaccurate measurements, and unnecessary maintenance costs. Without cross-domain analysis, misalignment between demand and asset design remains obscured, increasing long-term operational risk.
The frontend is a static web application built with SvelteKit. It provides a lightweight, reactive user interface that allows stakeholders to interact with the system through natural language queries. The frontend is statically rendered and served via a CDN, ensuring fast load times and minimal operational overhead. Its responsibility is limited to user interaction, authentication context, and request orchestration, keeping business logic out of the client.The backend is implemented in FastAPI and deployed on Cloud Run. This provides a fully serverless execution model with automatic scaling, request-based billing, and strong isolation. FastAPI handles request validation, authentication, telemetry queries, and AI orchestration through well-defined REST endpoints.All data is centralized in BigQuery, which acts as both the analytical engine and the vector store. Three primary data categories are managed:Time-series telemetry from pumps, compressors, and sensorsEngineering curves and operational envelopesEquipment datasheets and technical documentationStructured telemetry remains queryable via SQL, while curves and datasheets are processed into vector embeddings and stored alongside metadata. This enables hybrid analytical and semantic retrieval within the same platform.The AI reasoning layer is orchestrated using LangChain. LangChain manages prompt construction, tool calling, retrieval-augmented generation (RAG), and context assembly. For each user query, the system dynamically retrieves relevant telemetry slices and semantically matched technical documents, then synthesizes an answer grounded in both operational data and engineering specifications.This architecture cleanly separates concerns, scales elastically with demand, and supports extensibility across additional assets, models, and infrastructure layers while maintaining deterministic, auditable AI behavior.
The project produced measurable results across financial impact, operational efficiency, and decision quality.Financial Impact (ROI)The most significant outcome was the identification and quantification of approximately €1 million per year in unmeasured distributed gas. By correlating infrastructure telemetry with equipment specifications and operating envelopes, the system exposed systemic volumetric measurement inaccuracies that were previously invisible. This transformed an unquantified operational issue into a clear financial KPI, enabling targeted remediation and improved revenue capture.Operational Efficiency KPIsThe AI chatbot significantly reduced the time required to analyze infrastructure behavior. Engineering investigations that previously required days or weeks of manual data extraction, curve analysis, and expert interpretation were reduced to on-demand, query-driven analysis. This improved time-to-insight and reduced decision latency for both operational and strategic questions.Asset Performance & Measurement QualityThe system enabled infrastructure-wide detection of mismatches between actual operating conditions and design intent, such as assets consistently operating outside optimal ranges. This improved visibility into measurement reliability and asset utilization, supporting more accurate performance assessment and operational tuning.Decision Quality & ScalabilityBy embedding engineering reasoning into an AI-assisted interface, the project reduced reliance on individual subject-matter experts and increased consistency in decision-making. Stakeholders across engineering, operations, and management gained access to a shared, technically grounded view of infrastructure performance.Strategic ValueBeyond immediate financial impact, the project established a scalable analytical foundation. The same architecture can be extended to additional assets, regions, or optimization use cases with minimal incremental cost.
This project demonstrates technical innovation, measurable impact, and operational excellence, making it a strong candidate for recognition.Technically, it integrates complex, heterogeneous datasets—real-time telemetry, engineering datasheets, and operational curves—into a unified lakehouse architecture. By vectorizing and semantically embedding design and operational data, the system enables retrieval-augmented AI reasoning at scale. The combination of FastAPI backend on Cloud Run, SvelteKit frontend, BigQuery storage, and LangChain orchestration reflects a modern, maintainable, and cloud-native solution that balances performance, scalability, and reliability.The project delivers tangible, quantifiable results. Across the entire gas network, it identified and quantified volumetric measurement inaccuracies worth approximately €1 million annually. It reduced analysis time from weeks to near real-time, improved decision-making confidence, and provided a scalable foundation for future operational and capital planning.Moreover, the project exemplifies innovation in human-AI collaboration. By embedding engineering reasoning into an AI-powered chatbot, it democratizes complex operational intelligence, reducing reliance on individual experts and enabling faster, consistent, and auditable decisions.Finally, the achievement is notable for a small, two-person team executing end-to-end development in six months, delivering production-grade software with immediate financial, operational, and strategic impact. This combination of technical rigor, measurable outcomes, and efficient execution underscores why the project represents excellence.
.svg){kind=icon}