A property search box doesn't need an LLM
How Hut Yangu turns a renter's plain-language query into filters, ranks the matches and widens the search when needed.
People do not think in filter forms. They think: “2 bedroom in Kilimani under 30k.” On a phone, selecting each field one by one can make a simple search feel like paperwork.
For Hut Yangu, I wanted a renter to write that sentence and get useful homes back. The search turns the query into filters for bedrooms, area and budget, ranks the results, and widens the search step by step when nothing matches.
Use the smallest tool that fits
The query is natural language, but the job is bounded. We know the fields people need to search and the kinds of values those fields accept. For that problem, rules are faster, cheaper and more predictable than calling an LLM.
That choice also makes the behavior easier to reason about. If a query says “2 bedroom,” the parser should capture a bedroom count. If it includes a neighborhood and a budget, those should become filters too. We can test those transformations with concrete examples instead of hoping a model interprets them the same way twice.
No search system can promise a match for every combination. If strict filters return nothing, stopping at an empty page is not always helpful. Hut Yangu widens the search in steps, so renters can discover nearby options instead of having to guess which filter to remove.
Plain language still needs product decisions
The parser is only one part of the experience. Ranking decides which homes appear first, and widening rules decide what to relax when the exact query has no results. Those are product choices. A clever model cannot make them on the team’s behalf.
Rules do have limits. This search understands a known set of property fields; it is not a general assistant. That is a good trade for a focused search box. If the product later needs open-ended questions, that would be a different job with different boundaries.
Hut Yangu is pre-launch, with the web and mobile apps working end to end in development and a waitlist open. The point of this search is simple: meet renters where they are, then make the path from a sentence to a useful set of homes short and predictable.