Housmart: The AI-Native Real Estate Investment Decision Engine

  • Vision: Our vision is to turn fragmented data into strategic decisions

  • Role: Co-Founder and Lead Product Manager

Who are the users?

  • Real estate investors and agents

Problem & VoC

Investors struggle to evaluate properties because many location factors (safety, amenities, schools, neighborhood demographics) are fragmented and hard to interpret. Visiting 5-6 different platforms to gather data is time-consuming (taking hours, even days per property), prone to human bias, and difficult to scale, often leading to "analysis paralysis."

Data-driven feature prioritization for MVP is based on 50+ surveys.

The Solution: An AI Hypothesis

If Housmart delivers AI results in “transparent investment scoring + narrative explainability”, with "minutes latency and adaptive recommendations” to real‑estate investors/brokers, he/she is in position to quickly understand risk/ROI/suitability and make decision confidently without manual search or needing an analyst, which in turn creates higher user trust and repeat usage, which Housmart can capture a percentage in subscription revenue and premium “pro insights” upsells.

I built HouSmart MVP to function as an intelligent analyst. The platform aggregates disparate data sources and uses Generative AI to synthesize them into actionable investment strategies.

  • Aggregation: Automatically fetches real-time market data.

  • Synthesis: leverages Google's Gemini LLM not just to summarize data, but to reason through it-generating an objective "Investment Score" (0-100) based on cash flow and appreciation potential.

  • Personalization: Unlike static reports, HouSmart remembers user preferences via Supabase, delivering tailored recommendations.

User Journey Map:

This journey map illustrates the end-to-end experience of our user.

Vibe-Coding: Proof of Concept

I built HouSmart MVP using Google Antigravity Coding tool as an intelligent analyst. The platform aggregates disparate data sources and uses Generative AI to synthesize them into actionable investment strategies.

  • Aggregation: Automatically fetches real-time market data.

  • Synthesis: leverages Google's Gemini LLM not just to summarize data, but to reason through it-generating an objective "Investment Score" (0-100) based on cash flow and appreciation potential.

  • Personalization: Unlike static reports, HouSmart remembers user preferences via Supabase, delivering tailored recommendations.

Team Leadership: Full-stacked product

Then I started to work with a team of developers with the prototype and PRD for a full-developed product with key features:

  • Customized AI Insight Score: A proprietary scoring algorithm that evaluates location quality instantly, color-coded for quick decision-making.

  • Customized Narrative Synthesis: Leverages LLM not just to summarize data, but to reason through it, generating an tailored "Investment Recommendation" based on individual user’s background and local policy, by integrating the RAG.

  • Adaptive Feedback Loop: A natural language feedback system where the AI "learns" from user critiques (e.g., "This area is too noisy"), updating the user's profile in the vector database to refine future analyses.

Model Selection:

Selecting the optimal AI model is crucial for Housmart's success. We carefully evaluate potential models based on several key criteria to ensure we deliver the most accurate, reliable, and user-friendly experience possible. Here's what matters most:

AI Architecture and Tech Stack

Road Map and GTM

Impact & Results

  • Efficiency: Reduced the time required for comprehensive property analysis from >5 hours to <3 minutes (a >99% reduction). Empowered investors to screen dozens of properties daily, allowing them to focus physical site visits only on high-scoring opportunities.

  • Execution: Successfully built the prototype via vibe-coding in 2 weeks and led a team of developers to deliver a fully functional product from PRD to production in just 3 months, demonstrating end-to-end product leadership and technical execution.

80+ 95% 55%

Active Beta User Analysis Completion Stated Willingness to Pay (WTP)