Teaching an Ontology to Read Highway Engineering Documents: A Semantic AI Metadata Platform for Washington State DOT.

About the Client

WSDOT is the state agency responsible for building, maintaining, and operating Washington’s multimodal transportation system, including its highways, ferries, and rail corridors. The department is responsible for more than 20,000 lane-miles of roadway and nearly 3,000 vehicular bridges, and operates on a biennial budget exceeding $11 billion. An agency of that scale generates a continuously expanding body of engineering knowledge — design manuals, specifications, environmental documentation, and project records — that engineers must be able to find quickly and trust completely.

Industry Context

State departments of transportation share a common challenge: they are among the most document-intensive engineering organizations in government, yet most were built on content systems designed for storage and version control rather than for helping engineers find the right concept inside thousands of technical documents. The NCHRP, a nationally pooled research program funded by state DOTs and administered through AASHTO and the Transportation Research Board, had already published NCHRP Report 846, Improving Findability and Relevance of Transportation Information, laying out the problem. WSDOT’s engagement with Iknow and Spy Pond Partners was the applied, technology-integration phase of that research, later documented alongside parallel state DOT pilots in NCHRP Research Report 947. Commercial content platforms used by public infrastructure agencies are generally strong at storage and version control, but few include a semantic layer capable of understanding that a culvert is a kind of drainage asset — the gap WSDOT asked Iknow to close.

Current Situation

WSDOT already used the Drupal content management system, backed by a MySQL database, to store, present, and manage its engineering knowledge, procedures, and specifications. What it lacked was a way to enrich that content with concept-level metadata at scale — the foundation needed to move beyond keyword search toward genuine precision and recall. WSDOT contracted with Iknow, through prime contractor Spy Pond Partners, to integrate software capable of autoclassifying engineering documents as they were ingested.

• Engineering content that plain-text search could not adequately serve. WSDOT’s Drupal-based library held thousands of documents, but conventional keyword search could not capture relationships between engineering concepts — for example, that a culvert is a kind of drainage asset — so relevant results were easy to miss.
• No scalable way to apply metadata. Manually tagging WSDOT’s existing and continuously growing document library with consistent, meaningful metadata was not realistic given the department’s staffing and budget constraints.
• A solution that had to work within, not replace, existing systems. Any platform needed to integrate with WSDOT’s existing Drupal and MySQL environment rather than requiring a costly, disruptive replacement — a real constraint for a publicly funded agency.
• An unproven approach that needed to be demonstrated, not assumed. Because the premise — that content enrichment and computational classification could improve findability — was itself the subject of the research, the engagement had to produce evidence, not just a working system.

• Demonstrate that content enrichment and computational classification could improve navigation and search precision and recall.
• Integrate software components into WSDOT’s existing Drupal and MySQL environment to autoclassify documents as they were ingested.
• Build a domain-specific ontology describing the semantic and lexical relationships among WSDOT’s engineering concepts.
• Validate the resulting platform against WSDOT’s document library and confirm accuracy with the engineers who would rely on it.

How Iknow Structured the Work

Iknow designed and integrated a seven-component metadata processing platform that layered semantic AI capabilities on top of WSDOT’s existing Drupal and MySQL environment rather than replacing it, orchestrated by a custom-built integration component, Taggr.

Key Activities & Decisions

• Ontology development. Using Protégé, an open-source platform for building knowledge domain models, Iknow constructed an ontology describing the semantic relationships among WSDOT’s engineering concepts, built from a taxonomy of parent-child class relationships.
• Knowledge graph and semantic extraction. Iknow deployed Apache Jena to apply that ontology to WSDOT’s document corpus, extracting concept relationships as semantic triples — for instance, that a culvert is a kind of drainage asset — and storing them in a triple store built for interconnected semantic data.
• Natural language preprocessing. Iknow incorporated Google Cloud NLP to lemmatize document text, score terms for salience, and strip out low-value words, producing a clean, high-signal list of concepts per document.
• Custom integration layer. Iknow built Taggr as the platform’s connective tissue: it normalized text case, routed content to the NLP engine, matched high-salience terms against the ontology’s triple store, and wrote tags into WSDOT’s Drupal content forms.
• Enterprise search integration. Iknow connected the newly tagged content to Apache Solr, an open-source search platform, so engineers could retrieve documents using the concept relationships the platform established.
• Validation testing. Iknow ran the platform against WSDOT’s document corpus across multiple rounds of testing, successfully classifying more than 2,000 documents and confirming results with WSDOT engineers.

Stakeholders & Collaboration

Iknow worked as a subcontractor to prime research contractor Spy Pond Partners within an NCHRP-sponsored, AASHTO-overseen research project, coordinating directly with WSDOT’s engineering and IT staff to ground the platform in real content and validate its results.

Introducing Semantic AI Into a Legacy, Resource-Constrained Environment

A ground-up semantic technology stack risked being too costly and disruptive for a public agency’s existing systems and budget. Iknow addressed this by building most of the platform — six of seven components — on freely available, open-source software, and by designing Taggr as a thin integration layer that worked with WSDOT’s existing Drupal and MySQL environment rather than requiring its replacement.

Achieving Genuine Engineering-Domain Accuracy

An off-the-shelf classification model would not understand WSDOT’s specific engineering vocabulary well enough to produce trustworthy results. Iknow addressed this by purpose-building an ontology around WSDOT’s own engineering concepts and validating the platform’s output against WSDOT’s document library across multiple rounds of testing with WSDOT’s engineers, rather than treating a single test pass as sufficient proof.

Quantitative Outcomes

• Documents classified: 2,000+ engineering documents, correctly autoclassified across multiple rounds of testing.
• Technology components integrated: seven, including six open-source tools and one custom-built integration module (Taggr).
• Search accuracy: confirmed by WSDOT engineers as meeting the department’s expectations for precision and recall.

Qualitative Outcomes

WSDOT’s engineers gained the ability to search engineering content by underlying concept rather than exact keyword — finding, for example, all documents related to a class of drainage asset even when individual documents used different terminology. The ontology-driven knowledge graph Iknow built anticipated an approach that has since become mainstream, as many organizations now pursue similar semantic, retrieval-based architectures using large language models. Because the engagement was an NCHRP research pilot, its methodology and results were documented in NCHRP Research Report 947, published for every AASHTO member state DOT.

Timeline to Impact

Iknow delivered a validated, working platform within the engagement’s roughly 20-month period of performance, with accuracy confirmed against WSDOT’s own content and engineers before the research concluded. The findings were published nationally in 2020, giving other state DOTs an evidence-based model to use immediately.

Core Skills

• Metadata & taxonomy strategy
• Semantic AI / ontology development
• Enterprise search architecture
• Legacy systems integration

Methods & Frameworks

• Ontology and knowledge-graph engineering
• NLP-based content enrichment
• Phased platform integration and validation testing

Technologies & Tools

• Drupal, MySQL, and Protégé
• Apache Jena, Apache Solr, and Google Cloud NLP
• Custom-built Taggr integration module