A single view, not a single source

The single truth is often complicated.

gives you the insights needed to draw your own qualified conclusion.

How many times have you experienced a Single Source of Truth in your construction projects?

provides a Single View of Truth that spans across the knowledge repositories provided.

Links, not lock-in

Day 1's app-centric trap was "which application should own this data?" The answer turned out to be: none of them have to.

Learnings and perspectives

• Our Knowledge Graph establishes loose connections between files on different systems
• Not only documents but the fragments within them are linked
• The graph sits on top of existing systems and can be implemented in a non-intrusive way
• The quest of discovery always has a starting point. A type of object, a geo-location, a material, a person or company. The more context we establish between fragments, the better discoverability we can deliver
• We deliver the processing pipelines and expose the result with our agent

Nobody has to change how they work

• The DMS keeps being the DMS
• The ERP keeps being the ERP
• The architects keep using the tools they already argue about

This is why it can actually be adopted.

Provenance, revisited

Remember the door

Two contradictory facts. Nothing deleted. And once we knew who said it, when, from what source, with what confidence, the contradiction became answerable.

Every fact in the graph can carry its own history

Which is what lets a person — or an agent — reason about which version of a contested fact to trust, in a given context, rather than pretending contradictions don't exist.

• Same entity represented in multiple contexts
• Different visual representaitions of the same entity
• Different historical versions of the same entity

04 — Cue Agent

Ask questions. Get answers. With sources.

The Cue Agent is an AI assistant purpose-built for project knowledge

It uses dedicated tools to search, query and reason across the full index — combining your project knowledge with publicly available information

Every answer traces back to a source document

AI agents as the new front door

Instead of learning a query language or navigating five separate tools, you ask a plain-English question. The agent translates it, runs it, and hands back an answer with its sources.

MCP

The index is accessible from anywhere via MCP and a CLI

MCP — the Model Context Protocol — is an open standard that lets AI assistants connect directly to a live data source and query it in natural language. No custom integration code. No query language to learn.

Once connected, "ask the graph a question" becomes as ordinary as asking Claude anything else.

From documents to actionable knowledge

The whole arc of the course, on one line.

A single question, three stakeholders

"How many windows are on the northern façade, are any past their maintenance interval, and does the current lease allow us to replace them?"

In a graph built on BOT and an Object Type Library, it's one traversal

Spatial topology, maintenance records and contract terms all resolve to the same canonical windows — queryable by anyone, answerable by an agent, with every claim traceable to its source document.

Is this the right fit?

Honest answers

A knowledge graph is not the right choice when…

• You have a single, well-defined data source already available in JSON or a database — a direct API is simpler and faster
• Your use case is a standard CRUD app with no need to reason across documents or entities
• You need results in days and the data is already clean, structured and complete
• You want to extract a few entities from a handful of documents — a simple LLM call or a lightweight script will get you there faster and cheaper

It is the right choice when…

• Knowledge is scattered across hundreds of files — drawings, models, contracts, reports — and you need to work across all of it
• You need to connect entities across different sources — the same asset appears in an IFC model, a P&ID, a contract and a maintenance log
• You want to build apps or run agents on top of the whole landscape of project knowledge, not just a slice of it
• Traceability matters — every answer must be backed by a source document
• Your data volumes, document diversity or cross-referencing needs are beyond what a hand-rolled solution can realistically handle

The signal: are you dealing with the full complexity of a large project?

If yes — that is exactly what this was designed for

Capstone Part A: "Draft your Object Type Library"

Always run this first. One sheet of paper per group.

Relationships, written as sentences

If it reads like a sentence, it will work as an edge.

The blank grid

Eight to ten cards. Business language, not ontology language.

Closing case by cohort

🖥️ Hands-on lab, Part B

"Ask it for real"

Requires a Cue sandbox project and the MCP connector

Questions worth asking

The point isn't the sophistication of the question.

Key takeaway

"This isn't a new system I have to convince everyone to adopt — it's the layer of shared meaning that finally lets everything we already have talk to each other, and to AI."

"And I just watched my own spreadsheet become part of that — no code, no IT ticket, no waiting."

Six days, six ideas

• 1 — Our data already behaves like a graph
• 2 — A fact is a claim, not a truth
• 3 — New questions don't need a new schema
• 4 — Ten simple ideas describe any building
• 5 — One shared vocabulary, not one system
• 6 — Everything we have can now talk to each other

What to do on Monday

• Take your Object Type Library sheet back with you
• Find the one authoritative list your organization already has — the asset register, the room programme, the member directory
• Pick the one question that spans two systems, and use it as the business case

Start with one question. Not with a platform.

Thank you — and good luck with your first Object Type Library.