Projects
Volve.Evaluate web app development
Applying LLM-Powered AI to Tender Document Analysis for Construction and Real Estate Projects

Design and development of an MVP application that facilitates deep document analysis and bid comparison for tendering purposes.

Customer

Volve

Maturity

Start-ups

Industry

Construction, Real Estate

Country

Norway

Start

2024

Duration

ongoing

Team size

5 people

Applying LLM-Powered AI to Tender Document Analysis for Construction and Real Estate Projects by zaven

Technologies

Services

Generative AI Software Development

MVP Software Development

Front-end Software Development

Back-end Software Development

DevOps as a Service

Agile Software Development

Goal

Volve.Evaluate approached Zaven with a vision to transform tender document analysis for the construction and real estate sectors. They sought an intuitive product that would seamlessly integrate into the workflows of tender analysts, making bid evaluations more efficient and insightful. Key objectives included the development of a robust data model and AI-powered prompting system to extract and organize bidding data effectively. The project timeline was ambitious: deliver a proof-of-concept web application within two months and a fully functional MVP within four months to attract further investment.

Challenge

A primary challenge was ensuring the factual accuracy of AI-generated outputs, as any inaccuracies could compromise the reliability of tender analysis. Additionally, the solution needed to provide precise citations with source quotes, regardless of the original document's format, to maintain trust and transparency.

Processing diverse user-uploaded documents, including PDFs, required a balance between efficiency and cost-effectiveness. Leveraging LLMs for document analysis presented high operational costs, demanding innovative approaches to keep token usage under control.

Developing a user-friendly interface was critical to adoption. The interface needed to present AI-generated data in a clear and intuitive way, allowing users to easily understand the context and trace information back to its sources.

Solution

Zaven employed a modern Retrieval Augmented Generation (RAG) architecture for the MVP. This cloud-native solution leveraged Azure AI Search and Azure OpenAI to process documents and generate structured insights. To future-proof the system, a flexible solution architecture was designed, capable of replacing the basic RAG with a Python-based dsRAG mechanism, which could integrate multiple LLM models and vector databases for enhanced customization.

To improve semantic search accuracy, semantic chunking was implemented, ensuring contextually rich document segmentation. For data extraction, Microsoft Document Intelligence handled complex documents, while lightweight Python scripts were used for simpler cases, balancing performance and cost.

The user interface was designed with tender analysts in mind, combining clarity and ease of use. Drawing from familiar patterns in AI applications, the UI presented tender packages and bid relationships intuitively. Source traceability, visual hierarchy, and actionable insights were seamlessly integrated, ensuring the interface was both user-friendly and powerful.

Outcome

Zaven delivered a robust MVP underpinned by a scalable cloud infrastructure and a streamlined CI/CD process. The application featured a brand-aligned UI built with Bootstrap, designed to simplify workflows and enhance user experience.

The front end, built in React, was seamlessly integrated with a .NET-based back end hosted on Azure's cloud-native infrastructure. The system also leveraged OpenAI's ChatGPT for advanced AI capabilities, empowering tender analysts with accurate document analysis and bid comparison tools.

With its RAG-based architecture, the MVP successfully addressed factuality, citation traceability, and cost efficiency, ensuring a reliable and scalable solution. This project set a strong foundation for future growth, equipping Volve.Evaluate with an innovative tool to revolutionize tender analysis processes.

From Hype to Hard Hats: Practical Use Cases for AI chatbots in Construction and Proptech.

Remember the multimedia craze in the early 2000s? It was everywhere, but did it truly revolutionize our lives? Probably not. Today, it feels like every piece of software is labeled “AI-powered.” It’s easy to dismiss AI chatbots in construction as just another tech fad. But here’s the thing: AI in construction isn’t about flashy gimmicks....

Are you looking to make something similar?Let's talk!

Zaven built us a powerful AI Solution that validated our concept. The positive response from our users has been great, and we're now scaling the solution with their help. I'm very pleased with their flexibility and commitment to our success.

Herman Smith (CEO & Co-founder at Volve)

client

Volve (Norway)

Volve is a Norwegian technology company that combines industry expertise in real estate and construction with artificial intelligence to empower decision-makers to optimize cost, time, and sustainability in project execution.