Smart Retrofit Solutions — Using BIM and AI for Smart Building Upgrades
Most existing buildings have no digital plans, no automated code check, and no easy way to plan energy upgrades. We built the system that changes that.
80% of buildings that will exist in 2050 are already standing today — and most are energy-inefficient, potentially non-compliant with current codes, and lacking the digital records needed to plan upgrades. The building sector accounts for nearly 30% of global energy consumption. Retrofitting is essential, but the current process is manual, slow, and error-prone. Engineers spend days manually modeling buildings, checking codes by hand, and running simulations in disconnected tools. No single platform existed to do all three together. That was our problem to solve.
Our case study — a commercial building at 10225 Whalley Blvd #102 in Surrey, BC, treated as an existing building without complete design plans to demonstrate the full retrofit workflow. We received a point cloud data donation of the bottom commercial part of the building from McElhanney which was used for the project.
Alignment With UN Goals
Our solution is a three-phase approach.
PHASE 1 — Point Cloud to BIM Model
Imagine trying to renovate a building with no blueprints. A laser scanner was used to capture millions of data points representing the building's exterior — then we cleaned, aligned, and converted that data into a full 3D digital model.
.png)
.png)
.png)
The misaligned LiDAR dataset received for our project was cleaned and aligned in CloudCompare as shown in the left photo, indexed in Autodesk ReCap as shown in the middle photo, and reconstructed as a full BIM model in Autodesk Revit as shown in the right photo. The process involved removing vegetation and terrain noise, spatially registering the scans, and manually modeling the building envelope, walls, roof, and openings, using the point cloud as a geometric reference.
PHASE 2 — AI Building Code Compliance
Normally, an engineer spends hours reading building code documents and manually cross-checking them against a model. Our plug-in does this in seconds — ask it a question, get a cited answer, and see every violation highlighted in red directly inside the model.
A Revit plug-in built on a Retrieval-Augmented Generation (RAG) pipeline using Anthropic Claude Haiku 4.5. The plug-in highlights compliant elements in green and violations in red directly inside Autodesk Revit, with no manual checking required. It can also identify whether a building is residential or commercial. Shown in the figures above, the system is checking the case study which is a commercial model which uses the British Columbia Building Code (BCBC) 2024 Part 3, which covers larger, more complex structures. For this project, we showed an example checking door width and height.
We also designed the plug-in so it can run the building code check on a residential building. The system checks BCBC 2024 Part 4 — Housing and Small Buildings.
The built-in AI assistant was a design iteration we included that answers plain-language building code questions and cites the exact page of BC Building Code 2024.
Another iteration was that we designed the non-compliant element to highlight in red so designers can see specifically which element it is.
PHASE 3 — Energy Optimization
Flagging violations is step one — but the real opportunity is figuring out which upgrades deliver the greatest energy and carbon savings. So we built an optimization engine.
We ran energy simulations across 122 building configurations and used Pareto optimization to find the designs with the best trade-offs between total carbon, energy use intensity, and peak load. We also clustered by material type. Concrete and drywall buildings have the highest energy intensity, making them prime targets for upgrades, while wood has the lowest carbon footprint. That gives designers a clear starting point.
This chart shows carbon emissions broken down by material type across our dataset. As you can see, drywall and wood combinations account for the largest share of embodied carbon, at nearly twice that of concrete. Wood-only buildings demonstrated the lowest carbon impact. This finding aligns with existing literature supporting timber-based construction for lower embodied carbon.
In total, we analyzed 122 building design cases. Shown in the figure on the right, most frequent materials in the dataset were concrete at 32% and wood at 27%. We classified 6 material types overall: concrete, wood, drywall and wood, metal, brick, and unknown. This distribution gave us enough variability to demonstrate the optimization framework.
Overall System:
Smart Retrofit Decision Making for Existing Buildings
The end result is an integrated decision-support tool: engineers get a code-compliant, energy-optimized retrofit recommendation for an existing building, starting from nothing but a LiDAR scan
Capstone Presentation Video (5min)
Extra Information:
Key Design Iterations:
Phase 1 — Our original plan was to automate the point cloud conversion to BIM model, however, in the beginning we were given a very challenging residential point cloud model. During the third client meeting around January 8th, it was decided that the current point cloud dataset was large and difficult to work with due to loading and merging challenges. Therefore, it was decided that a different dataset from McElhanney would be used, obtained through a data donation, as it would be more realistic to work with for the project. The donated dataset, Whalley Blvd., was received by the team on January 26th. Due to the point cloud data being received later in the term, the client decided not to automate the point cloud conversion due to time constraints and instead focus only on converting the point cloud data into a Revit model that could be used for Phases 2 and 3.
Phase 2 — We initially planned to generate Python scripts for each compliance check. After client consultation, we switched to a JSON rule card architecture making rules reusable across any building, traceable to specific code pages, and far more reliable. A chat assistant and visual violation highlighting were added as final design iterations, both fully delivered.
Phase 3 — This phase was originally scoped to include cost optimization alongside energy and carbon. Mid-project, reliable cost data proved unavailable and the complexity exceeded our timeline. After client discussion, we decided to focus exclusively on energy and carbon.
Contirbutions:
Phase 1 — Kasra in civil engineering worked on this phase converting the point cloud to a useable Revit model for phases 2 and 3.
Phase 2 — Colby and Jaeda worked on this phase. Colby was mainly responsible for the coding and computer aspect of the project, while I helped with the civil engineering aspect specifically with BIM, Revit, and BCBC rules.
Phase 3 — Kurt, Daniel, and Arshia worked on this phase, ranging from civil, mechanical and electrical engineering.