Zeme is an end to end rental marketplace that enables renters to search, apply, and complete rental transactions on a single platform. It serves hundreds of thousands of users and has closed 4,000 to 5,000 apartments. The platform already supported real time listings and had a basic AI chatbot prototype for property search using natural language queries.
Problem: traditional search limits AI property search adoption
Rental platforms rely on filters like price, location, and amenities, which limit how users express real needs such as commute time, nearby schools, or access to multiple locations. Even though Zeme had an initial AI chatbot, it lacked strong search intelligence and result optimization. This led to a heavy dependence on manual, filter based search instead of an effective AI driven experience.
The system struggled to handle multi factor queries, had weak context awareness when users refined their inputs, and offered only basic geolocation without deeper transit intelligence. In addition, there was a weak connection between what users asked in the chatbot and the actual search results shown. Because of these gaps, users experienced friction during property discovery, which reduced the overall effectiveness and adoption of AI property search.
Solution: conversational AI property search assistant for intelligent discovery
GoML enhanced Zeme’s existing AI chatbot by improving how queries are understood, processed, and converted into results. The system enables natural language search with relevant property recommendations, using LLMs, geospatial intelligence, and an Agentic AI Accelerator to manage context and orchestrate search workflows.
Conversational AI and context management
Example query: Apartments under 2000 dollars within 30 minutes of office and near good schools
Key improvements:
- Strong handling of multi factor queries
- Better interpretation of user intent
- Context maintained across multiple interactions
- Query refinement without restarting the search
- Smooth conversational experience for discovery
Geolocation and route intelligence
The system integrates location intelligence to improve accuracy:
- Travel time calculation using Google Maps APIs
- Neighborhood aware property matching
- Multi location commute optimization
- POI based ranking with walking distance priority
Search orchestration and ranking
The enhanced AI property search combines multiple systems:
- Property database queries from MySQL
- Semantic search using Qdrant vector database
- Redis caching for faster geolocation queries
Interactive user experience
The output is not limited to a map interface. The system automatically converts user queries into structured filters and applies them in real time.
Key experience improvements:
- Automatic filter application based on natural language input
- Results shown in both map view and property listing panel
- Visual property markers with contextual insights
- Real time updates based on user queries
- Seamless interaction between chat, filters, and results
This creates a faster and more intuitive property discovery experience.
Infrastructure and deployment
The system uses a scalable cloud stack:
- LLMs via Amazon Bedrock
- Backend APIs using FastAPI
- AWS Lambda and EventBridge for processing
- S3, Qdrant, and Redis for data storage
- CloudWatch for monitoring
Quality assurance
Validation focused on accuracy and usability:
- Testing multi factor query handling
- Validating context switching
- Ensuring commute and POI accuracy
- End to end testing with real data
Impacts
- 30 to 40% higher accuracy for complex queries
- 2 to 3X faster property discovery
- 50% less reliance on manual filters
- Supports 3x to 5x user growth
- 25 to 35% better user engagement
About
Before Gen AI and after Gen AI
“With AI property search, Zeme transformed rental discovery into a faster, context aware experience that helps users find better homes efficiently.”
Prashanna Rao, Head of Engineering, GoML
Key takeaways for rental platforms
Common challenges
- Manual search does not scale
- Lack of context aware systems
- Difficulty handling complex queries
Practical guidance
- Adopt AI property search for natural language queries
- Combine geolocation and POI intelligence
- Maintain conversation context for better UX
- Start with high impact use cases like commute and amenities
Ready to build AI property search solutions
Partner with GoML to build scalable AI property search systems using Gen AI with AI Matic.
