Introduction:
Build a deployed voice-first AI property discovery assistant for Bengaluru that helps renters/buyers find properties based on spoken preferences, evaluates whether listings actually fit their lifestyle, explains why listings were shortlisted, allows voice-based refinement, and books a site visit.
The core product principle is:
"Don't just help users find properties. Help them understand whether a property actually fits their life."
AI Tools:
Stitch by Google
Cursor
n8n
AI-Powered Property Discovery & Shortlisting:
Designed a voice-first property discovery experience that helps users evaluate homes based on their lifestyle, budget, commute, transit access, and essential amenities, not just listing details.
The experience uses real property listings and location data to enrich search results, rank properties against user preferences, and explain why each property is shortlisted or excluded. Users can refine their shortlist conversationally, updating individual constraints without restarting the search or losing existing preferences.
Key UX focus areas-
-
Voice-first search and natural-language preference extraction
-
Contextual clarification and constraint confirmation
-
Real-data-driven property matching and ranking
-
Conversational shortlist refinement with persistent preferences
-
Transparent recommendations with evidence-based reasoning and neighborhood citations
Design goal-
Help users make more confident property decisions by turning property search from a listing-browsing task into a personalized, explainable decision-making experience.
The assistant should combine-
-
Real, currently available property listings
-
Real neighborhood and geospatial data
-
Voice-based conversational interaction
-
RAG-grounded explanations with citations
-
Transparent shortlist reasoning
-
Voice-based shortlist refinement
-
Site-visit booking
-
n8n automation to generate and email a PDF shortlist
Cursor — Define workflows:
Building the AI Architecture in Cursor: RAG, MCP & Data Cleaning
I used Cursor as my primary development environment to build and iterate on Property Scout, a voice-first AI property discovery experience. My focus was to make property recommendations grounded in real listing and location data rather than generic AI-generated assumptions.
Data cleaning and preparation: I worked on cleaning and normalizing property listing data, standardizing fields such as location, rent, bedrooms, and amenities, and removing incomplete or irrelevant records. I also accounted for data freshness and source traceability so recommendations could be linked back to actual listings.
RAG (Retrieval-Augmented Generation): I explored a retrieval-based architecture to ground neighborhood insights and property explanations in relevant external sources. The goal was to retrieve contextual information, such as neighborhood guides and practical location details, and use it to generate evidence-backed responses with citations instead of relying on the LLM's internal knowledge.
MCP server integration: I incorporated an OpenStreetMap MCP integration to retrieve nearby points of interest and transit information. This helped connect property listings with real-world context, supporting questions about metro access, nearby amenities, and location suitability. I also focused on making real MCP calls and avoiding fabricated locations or unsupported claims.
This project helped me connect my product design thinking with hands-on AI development. I gained a deeper understanding of how data quality, retrieval, tool integrations, and orchestration influence the reliability of AI-powered user experiences.
Local Evaluation Testing & Reporting
I ran local tests in Cursor to validate Property Scout AI before deployment, focusing on preference updates, budget and commute constraints, and evidence-backed recommendations. I also tested edge cases such as unintended changes, unavailable listings, and unsupported claims.
I built an automated evaluation report in Markdown to review results by category and identify potential failures.
Result: All 22 local evaluation tests passed across Edit Correctness, Feasibility, and Grounding & Hallucination.
Final UI — Refined Experience
I designed and prototyped Property Scout, a voice-first AI property discovery experience, using Stitch by Google for UI exploration and rapid prototyping, alongside Cursor for implementation and iteration.
The experience helps users find homes based on budget, location, lifestyle, and practical needs through conversational search, preference confirmation, and AI-powered property matching. I explored personalized shortlists with match scores, rent, transit proximity, and property highlights, enabling users to evaluate options and refine their preferences naturally.
My focus was to combine thoughtful UX design, real property data, and conversational AI to make property discovery more transparent, relevant, and easier to navigate.
→ View app
Project Demo Video
In this demo, I walk through Property Scout AI’s key features, from voice-based property search and personalized shortlists to real-world location insights, AI evaluation, and the n8n workflow powering the experience.
