Pioneering AI in construction

Executive summary: AI-powered innovation in the construction industry

01 Client

A leading Central European general contractor with more than two decades of experience and several hundred completed industrial, logistics, and commercial projects spanning millions of square meters. Recognized as one of the country's most innovative construction companies and a pioneer of digital transformation in the sector.

02 The core challenge

As a long-time adopter of BIM (Building Information Modeling) and Speckle as their Common Data Environment, the client wanted to take the next step in digitizing construction workflows by enabling project engineers, designers, and managers to interact with complex 3D models through natural language. Their biggest pain points were the time spent on repetitive querying of model data (such as material quantities, surfaces, or compliance checks), the inability to combine model intelligence with their existing tools, and the absence of a coherent vision for how AI should be embedded in the wider organization.

03 Solution delivered

Pretius delivered a two-part engagement: a working Proof of Concept (PoC) of an AI chat assistant embedded directly inside the Speckle App, and a long-term architectural vision for the client's future AI platform. The PoC is built around a modular multi-agent architecture orchestrated with CrewAI, communicating with Speckle through its GraphQL API, and integrated with Azure OpenAI as the language model provider. The architectural vision laid the foundations for scaling AI initiatives across the entire company.

Key outcomes:

A fully working PoC:
delivered as containerized applications (frontend microfrontend + Python backend), ready to run on the client's own infrastructure via Docker Compose.

A modular, agent-based architecture:
designed from day one to evolve from a single PoC into an enterprise-grade microservices platform.

A clear roadmap for the AI platform:
including high-level architecture diagrams, infrastructure proposals, and an optional monitoring vision.

Client background

The client is a leader in industrial construction and a pioneer of digital transformation in its region. The company has spent more than two decades building its country's industrial backbone. It has grown from a small branch of an international holding into a fully independent national enterprise that today serves as a general contractor of choice for hundreds of investors, including globally recognized brands across logistics, automotive, consumer goods, and industrial sectors.

The company specializes in the comprehensive delivery of industrial, logistics, commercial, and office facilities. With several hundred completed projects and millions of square meters of built space, the client holds a leadership position in its construction market. The company has been recognized with multiple national business awards, ranks among its country's leading enterprises, holds ISO 14001 and ISO 45001 certifications, and is a BREEAM-certifying body and member of its national green building council.

What distinguishes the client from the rest of the sector is its uncompromising approach to innovation. The company has positioned itself at the forefront of digital transformation in its country's construction sector, consistently developing the use of artificial intelligence and BIM methodology to plan investments with precision, optimize material usage, reduce risk, and maximize efficiency. This is exactly the mindset that led the client to look for a partner capable of moving them from BIM-as-a-data-source to BIM-as-an-intelligent-assistant.

The business challenge: From data-rich models to intelligent decisions

The client's BIM ecosystem revolves around Speckle – an open-source, object-based data platform for the AEC industry that serves as the company's Common Data Environment and connects models authored in Revit, Tekla, and other tools through a unified GraphQL API.

While Speckle already gave the company structured, queryable access to its 3D models, several issues stood in the way of unlocking the platform's full potential:

Time-consuming manual queries
Tasks such as calculating the total amount of concrete required for foundation footings, listing the masonry wall area on a given floor, or validating whether a model meets specific standards required engineers to either click through the model manually or write custom GraphQL queries – a skill that not every team member possesses.

Knowledge silos around GraphQL
Speckle's API is powerful, but its advanced use requires expertise. Without an abstraction layer, only a small group of specialists could fully benefit from the data already stored in the platform.

Many AI ideas, no clear roadmap
The client had a long list of high-impact AI ideas – from a chatbot for thousands of reference letters, through computer vision for counting elements on construction site photos, to construction progress overlays on 3D models. What was missing was a unifying architecture that would allow these initiatives to be built on a common foundation rather than as disconnected experiments.

Infrastructure ambiguity
With several existing tools already in the stack (identity management, ERP, BPM, workflow automation, an emerging data bus, and a document-based cognitive search), the company needed clarity on how an AI platform would fit into – and extend – this landscape, and what the trade-offs between on-premise and SaaS deployments would look like.

To turn Speckle from a model repository into a true intelligent assistant – and to lay the foundations for AI at scale across the company – the client needed a partner with both hands-on AI engineering experience and the architectural seniority to design a future-proof platform.

The solution provided by Pretius

A focused team of AI engineers and solution architects from Pretius worked in close cooperation with the client's BIM Product Owner and leadership team. After a series of workshops with the client's leadership, Pretius proposed a two-track engagement that would deliver tangible value in the short term while securing a sound architectural foundation for the long term:

#1

A Proof of Concept: an AI chat assistant inside Speckle

Pretius designed and built an AI-powered chat widget that integrates directly into the Speckle App as a microfrontend. The widget is embedded as an iframe with bidirectional communication, meaning it has full access to the user's session context – which project is open, which elements are selected, what the user is currently looking at – and can in turn drive the Speckle visualization, for instance by highlighting objects referenced in the conversation.

On the backend, Pretius built a modular Python application based on FastAPI, CrewAI, and the OpenAI SDK, and deployed it in Docker. The architecture is organized around the principles of Domain-Driven Design and Bounded Contexts, with each functional area encapsulated as an independent module:

Backend API (LLM Gateway)

The communication layer between the frontend and the AI modules.

Agents Orchestrator

The central component coordinating which agent or tool handles a given request, powered by CrewAI.

Domain Expert Agents

Deterministic, reproducible agents handling business-critical use cases (e.g., quantity take-offs, model validation, reporting).

Generic Speckle/GraphQL Agent

A general-purpose agent with full Speckle API documentation in context, capable of intelligently composing and executing read-only GraphQL queries.

LLM Proxy

A unified integration point with Azure OpenAI services.

Tools

A shared library of utilities such as calculators, analytical helpers, and external API integrations.

This separation between deterministic, narrow-scope "domain expert" agents and the broader "generic" agent reflects a careful balance: business-critical workflows demand predictable, reproducible behavior. At the same time, exploratory queries benefit from the flexibility of a general-purpose agent.

#2

An architectural vision for the future AI platform

In parallel with the PoC, Pretius's architects worked on a high-level vision for the client's enterprise AI platform. The deliverable includes logical architecture diagrams, descriptions of the core platform components, and a proposal for the supporting infrastructure. The vision is explicitly designed to host not only the Speckle assistant but also the client's broader AI ambitions – from RAG-based knowledge search over thousands of reference letters and legal documents, through computer vision for site photo analysis, to construction-progress overlays on BIM models.

Crucially, the architecture keeps modules and agents independent at the logical layer from day one. This means that, even though the PoC is deployed as two Docker containers running on a single virtual machine via Docker Compose, the same codebase can be progressively split into independent microservices and migrated to Kubernetes once the organization is ready – without a costly rewrite.

Measured outcomes and business impact

A pragmatic PoC with a strategic horizon

The chat assistant gives the client's teams an immediate, tangible way to interact with their Speckle models in natural language – asking questions like "What is the total amount of concrete required for all foundation footings?", "What is the masonry wall area on the first floor?" or "Does this model meet the required standards?" – without writing a single line of GraphQL.

Budget-optimal architecture, scale-ready by design

By keeping modules and agents independent at the logical layer and packaging the system as Docker containers, Pretius made it possible for the client to start on a single VM today and gradually evolve to a microservices deployment on Kubernetes tomorrow, with no architectural debt to clean up along the way.

A unified roadmap for AI across the organization

Beyond the PoC itself, the client now has a documented architectural vision that ties together the company's AI ambitions – from Speckle chat, through RAG-based document search, to computer vision and construction-progress visualization – on a shared technological foundation. Future AI projects will benefit from standardized communication protocols, shared infrastructure and reusable components rather than starting from scratch.

Why this engagement matters

This engagement is an example of how a focused, well-scoped AI Proof of Concept – when paired with a credible architectural vision – can act as a catalyst for enterprise-wide transformation. A few takeaways stand out:

Start narrow, design wide

A modular agent architecture lets you ship a focused PoC quickly while preserving the option to evolve into a full AI platform.

Treat AI as an organizational capability, not a feature.

A long backlog of AI ideas only becomes a coherent program when there is a shared platform underneath.

Combine deterministic and generative agents

Business-critical analyses need reproducibility; exploratory questions thrive on flexibility. Both belong in the same architecture, with clearly drawn boundaries.

Containerization buys you optionality

A Docker-Compose deployment on a VM today and a Kubernetes deployment tomorrow can – and should – share the same codebase.

Pretius demonstrated the ability to combine deep AI engineering expertise with the architectural seniority required to translate a CEO's vision and a CIO's roadmap into a single, coherent program of work in the construction industry.

Pioneering AI in construction: An agent-based chat for Speckle, built for a leading general contractor

Company

Country

Industry

System

Executive summary: AI-powered innovation in the construction industry

01

Client

02

The core challenge

03

Solution delivered

Key outcomes:

Client background

The business challenge: From data-rich models to intelligent decisions

The solution provided by Pretius

#1

A Proof of Concept: an AI chat assistant inside Speckle

Backend API (LLM Gateway)

Agents Orchestrator

Domain Expert Agents

Generic Speckle/GraphQL Agent

LLM Proxy

Tools

#2

An architectural vision for the future AI platform

Measured outcomes and business impact

A pragmatic PoC with a strategic horizon

Budget-optimal architecture, scale-ready by design

A unified roadmap for AI across the organization

Why this engagement matters

This engagement is an example of how a focused, well-scoped AI Proof of Concept – when paired with a credible architectural vision – can act as a catalyst for enterprise-wide transformation. A few takeaways stand out:

Start narrow, design wide

Treat AI as an organizational capability, not a feature.

Combine deterministic and generative agents

Containerization buys you optionality

Is your company ready to take the next step in AI-driven digital transformation?

Looking for a software development company?

Work with a team that already helped dozens of market leaders. Book a discovery call to see:

Michał Lipiński

Account Manager

We keep your data safe: ISO certified

Drop us a line at

Want to work with us?

What we do

How we do it

Products

Sitemap