Client background
Blue Lotus Properties is a real estate investment and brokerage company based in Seattle, Washington. The partners had identified an opportunity to leverage AI to improve how their team and clients evaluate properties — specifically, how AI could be used to surface insights from the unstructured data that flows through real estate transactions: property descriptions, inspection reports, neighbourhood data, comparable sales narratives, and client preference documentation.
The real estate industry has been slow to adopt AI, partly because the unstructured data challenge is genuinely difficult and partly because existing tools in the market provide generic outputs rather than insights tailored to specific investment criteria. Blue Lotus saw an opportunity to build an internal tool that could give their team a genuine analytical advantage — and potentially, if the concept proved out, a product they could offer to other real estate professionals.
The engagement was for a proof-of-concept iOS application: a purpose-built tool for the Blue Lotus team to evaluate AI-assisted property analysis workflows, gather internal feedback on which AI-powered features delivered real value versus which were noise, and demonstrate the capability to potential partners and investors.
The challenge
The challenge for Blue Lotus was not "can we build an AI product" but rather "what should an AI product in real estate actually do, and what does it need to do well to be genuinely useful rather than a gimmick?" This is a more sophisticated problem than it appears. The real estate AI landscape in 2024 was already populated with tools making broad claims about AI-powered insights — most of which delivered generic outputs that experienced real estate professionals found neither surprising nor actionable.
For the proof of concept to be valuable, it needed to demonstrate AI doing something that experienced real estate professionals couldn't do as quickly or as thoroughly themselves. The partners at Blue Lotus had specific ideas about where AI could genuinely add value: analysing inspection reports for non-obvious risk indicators, comparing neighbourhood data across multiple dimensions simultaneously, and generating structured summaries of properties relative to a specific investment thesis.
The iOS platform constraint was deliberate: the Blue Lotus team worked primarily on mobile, and a tool that required a desktop to access would not be used in practice. The proof of concept needed to be a genuinely mobile-first product, not a desktop product with a mobile-responsive layout.
Budget was a meaningful constraint at $10,000. This required the Squash Apps team to make disciplined scope decisions: defining a focused set of AI-powered features that could be built well within the budget rather than a broader set built superficially.
How we engaged
The engagement began with a one-week requirements workshop — conducted remotely — where the Squash Apps team worked with the Blue Lotus partners to define the specific AI use cases worth validating in the proof of concept. Three use cases were selected based on the combination of potential value and feasibility within the budget: AI-powered inspection report analysis (extracting and prioritising issues from unstructured inspection report text), property comparison (comparing multiple properties against a configurable investment criteria checklist), and neighbourhood insight synthesis (generating structured summaries from multiple data sources about a neighbourhood's investment characteristics).
The team for this engagement was lean by design: one senior iOS engineer with Swift and Core ML experience, one backend engineer for the AI integration layer, and a product designer for the UX work. The lean team structure was appropriate for a proof of concept — it kept communication overhead low and allowed for rapid iteration.
The partners at Blue Lotus were closely involved throughout the build, reviewing design prototypes and giving feedback on AI output quality. This feedback loop was particularly important for the AI features: getting the prompts and output structure right required several rounds of iteration based on real usage by experienced real estate professionals rather than testing with synthetic examples.
What we built
The iOS application delivered three core AI-powered features. The inspection report analysis module allowed users to photograph or upload a PDF inspection report and receive a structured analysis: issues grouped by severity and category (structural, mechanical, electrical, cosmetic), with a risk score for each category and a plain-language summary of the most significant concerns. The analysis was powered by a large language model via API, with carefully designed prompts that had been tuned through multiple rounds of iteration with the Blue Lotus partners to produce outputs matching how experienced real estate investors actually think about inspection reports.
The property comparison feature allowed users to define an investment thesis — desired return profile, risk tolerance, property type preferences, required characteristics — and compare multiple properties against that thesis simultaneously. The AI layer translated the natural-language investment criteria into a structured scoring model, rated each property against each criterion based on available property data and the user's notes, and presented a visual comparison that highlighted where properties diverged on the criteria that mattered most to the user.
The neighbourhood insight synthesis feature aggregated data from multiple sources (publicly available demographic data, crime statistics, school ratings, recent sales data) and used the language model to generate a structured narrative tailored to real estate investment decision-making — explicitly linking neighbourhood characteristics to the investment implications rather than simply presenting data.
The UX was designed for mobile-first use in the field: large touch targets, offline caching for property data already loaded, and a clean visual hierarchy that put AI-generated content in context rather than overwhelming the user with raw analysis output.
Technical approach
The application was built in Swift with SwiftUI for the interface layer. AI capabilities were integrated via API calls to a large language model provider, with the backend handling prompt construction, response parsing, and structured output extraction. A Node.js backend API served as the intermediary between the iOS app and the AI provider — handling authentication, caching of results, and the prompt management that was essential to output quality.
Prompt engineering was a significant part of the technical work. Achieving consistent, structured, high-quality outputs from a language model requires careful prompt design: specifying the output format, providing relevant context, and handling edge cases (e.g., inspection reports that are very brief or that don't follow standard formatting). The team developed a prompt library through iterative testing with real examples provided by the Blue Lotus team, documented the prompt design decisions for future maintainability, and built an output validation layer that caught malformed responses before they reached the user.
Document processing — particularly the ability to extract usable text from photographed inspection report PDFs — required combining iOS's Vision framework for image-to-text extraction with post-processing logic to handle the layout variations common in scanned documents. The accuracy of the final text extraction was high enough that the AI analysis downstream produced reliable outputs in the large majority of cases.
Results
The proof of concept was delivered in March 2024, on time and within the $10,000 budget. The Blue Lotus partners' assessment — shared in their Clutch review — was that the application demonstrated genuine AI capabilities in the real estate domain that they hadn't seen matched by existing tools in the market: "We were impressed by their technical abilities with AI-related concepts, they have deep understanding in this space."
The inspection report analysis feature was the most practically valuable in the partners' early usage: it surfaced non-obvious risk indicators in reports that the team would previously have read manually, saving time and improving analytical thoroughness. The neighbourhood insight synthesis feature required more iteration on the data sources and prompt design before it would be ready for regular use — a finding that informed the roadmap for the next phase of development.
The proof of concept achieved its primary goal: providing Blue Lotus with enough evidence to make an informed decision about whether to invest in building a full-featured version of the product. The partners are currently gathering customer data to validate which AI features their clients would find most valuable, with a view to a more complete product development engagement.
For real estate technology companies, investment firms, and PropTech startups exploring AI-powered property analysis, this engagement illustrates how large language model integration can surface actionable insight from the unstructured data that flows through every real estate transaction — inspection reports, market narratives, neighbourhood descriptions, and investment criteria documents. The key finding: achieving consistently useful AI outputs requires expert prompt engineering and iterative refinement with domain specialists, not just an API call to a language model. The Squash Apps team's depth in AI/ML integration delivered results that off-the-shelf solutions in the market were not matching.