All projects
data
Full-text search that ranks well
Build search over real documents and make the top result actually be the right one.
Why it matters
Search is everywhere and 'it returns something' is not the bar. Being able to measure relevance improvement is unusual and immediately credible.
What you build
- An index over a real document set
- Ranking better than naive keyword matching
- Typo tolerance and stemming
- A relevance evaluation you can rerun
If you want more
- Add faceted filtering
- Add semantic search with embeddings
Build it
- Add stemming and typo tolerance
Material for the whole build
Shows you can improve a fuzzy system by measurement rather than by opinion.
Get a plan built around projects like this