AI revolution in construction: weeks cut to minutes
Rite NRG partnered with Swedish founders to develop an AI-powered SaaS that reduces construction installation estimates from weeks to minutes. Through a three-month Discovery phase, we trained a custom AI model achieving 80% accuracy in recognising water pipes from 2D drawings – laying the foundation for a market-ready product.
Introduction
In the dynamic landscape of construction, traditional methods for cost and material estimation often lead to prolonged processes, potential errors, and significant manual effort. Two visionary investors with extensive experience identified a critical market gap: the weeks-long process of preparing installation offers and measurements could be optimised to mere minutes with the power of artificial intelligence (AI). The core problem was the highly manual, repetitive, and often tedious nature of performing take-offs – meticulously calculating pipe lengths and planning installation logistics from 2D drawings. They conceived a product, a web application operating as a SaaS (Software as a Service) solution in a subscription model, that would revolutionise this sector, starting in Sweden. Following success in Sweden, they confidently aim to promote the product globally. To transform this vision into reality, they sought a trusted technology partner and a fractional CTO to guide the technological direction, alongside a dedicated team to build the application.
The Challenge
The primary challenge was to develop an AI-powered solution capable of accurately analysing 2D architectural and technical drawings to generate precise cost and material calculations for all types of installations, starting with water installations. Initial tests with standard AI algorithms such as ChatGPT or Gemini proved insufficient for the specialised task of recognising installations on complex drawings.
The founders faced several significant challenges related to bringing their vision to life:
- Completing the founding team: The founding team, composed of a domain expert who also handled sales and an operations expert capable of building an operational structure, lacked a dedicated technical person, such as a CTO, who could provide strategic guidance on the technological direction of the product and complete their executive expertise.
- Unlocking effective product development: There was a significant gap in knowledge regarding how to build a scalable software product from conception to launch, including understanding the optimal processes and methodologies.
- Lack of development team: The absence of an in-house development team meant there was no workforce to develop and implement the AI solution. This also presented the dilemma of whether to undertake the costly and time-consuming process of building an in-house team from scratch, with inherent recruitment risks, or to leverage the proven expertise of an experienced nearshoring partner.
These factors highlighted the need for an external partner who could not only deliver technological expertise but also offer strategic guidance and ensure the product’s viability without incurring excessive initial costs.
Rite NRG's Solution
A strategic, phased approach was proposed, beginning with a focused “Discovery Phase.” This phase was designed to validate AI’s capability to support construction offer preparation and develop the core algorithm. The primary goal of Discovery was to teach an AI algorithm to recognise water pipes in drawings and achieve a specific quality level.
Key steps in our solution included:
- Initial feasibility study: Before committing to full development, Rite NRG conducted preliminary tests to assess if existing AI models (such as YOLOv8 and Detectron2) could effectively identify and segment technical installations on PDF drawings. This allowed for a cost-effective evaluation of the AI’s potential.
- Agile discovery phase: The Discovery phase was executed over three months, with weekly demos and status reports provided to the client, ensuring transparency and continuous feedback. This agile approach allowed for rapid adaptation and optimisation based on ongoing results.
- Product design workshop: As part of Rite NRG’s #riteway framework, we invited the client to a hands-on Product Design Workshop in Wrocław. This collaborative session, starting with breakfast and ending with dinner, fostered a strong partnership and allowed us to jointly define user needs, product features, and the minimum viable product (MVP) scope. This ensured that the technical development was perfectly aligned with the business goals and market strategy.
- Model Training: A crucial part of the Discovery phase involved preparing and processing technical drawings in PDF format, annotating images, and training the selected UNet + MobileNetV2 model. This iterative process aimed to achieve a high level of quality in identifying water installations.
- Overdelivery mindset: Rite NRG’s commitment to “overdelivery” meant going beyond just coding. We acted as a strategic partner, advising on the project’s direction and providing business insights to help the client build a profitable product and define its market strategy.
Technologies Used
The project leveraged Rite NRG’s expertise in AI, focusing on a robust tech stack for image recognition and data analysis. Our core framework was PyTorch 1.13.1 complemented by torchvision 0.14.1 and CUDA 11.6. For data processing, we utilised PIL/NumPy for image and mask manipulation, alongside a custom tiling system to efficiently handle large images (1024px tiles with 256px overlap). During the training and optimisation phases, we employed Adam optimiser with a ReduceLROnPlateau scheduler, using CrossEntropyLoss and Early stopping (patience=15). Initial model testing involved experimental libraries such as scikit-learn, pycocotools, and albumentations. After evaluating various models including Detectron2 (Mask R-CNN) and SegFormer, UNet with a MobileNetV2 encoder (ImageNet pre-trained) was selected as the optimal architecture, featuring a decoder with skip connections and an output of 6 segmentation classes (1024×1024 px). This setup proved most effective for accurate pipe segmentation in technical drawings.
Results
The Discovery phase was a resounding success, demonstrating Rite NRG’s ability to not only meet but exceed expectations:
- Trained Model: Within three months, the UNet + MobileNetV2 model was successfully trained, achieving an 80% quality level in recognising water installations, with clear potential for further improvement through continued algorithm training.
- Defined MVP Roadmap, Timeline, and Costs: Following the Product Design Workshop, a detailed MVP scope, effort estimation, timeline, and budget breakdown were prepared, providing the client with full clarity and confidence for the next steps.
- Strategic Partnership Established: Rite NRG proved to be more than just a vendor, acting as a strategic partner deeply invested in the client’s business success, aligning technical delivery with overarching business objectives.
This successful Discovery phase laid a solid foundation, proving that AI can indeed revolutionise construction cost estimation and providing a clear path forward for developing a groundbreaking SaaS solution.
Read more case studies
See how our teams deliver predictable results in complex environments.
Frequently asked #questions
Clear answers to the most common questions about our delivery model and approach.
Turn AI into real delivery impact
Your #Next-Gen-AI delivery starts here
As a nearshore delivery partner, we build and scale SaaS products using senior teams, strong culture fit, and AI to deliver up to 50% faster — without sacrificing quality or control.