Rekall is an open source tool to measure search relevance. This is highly inspired by Quepid. Quepid is one the most amazing open source tools out there to measure and tune search relevance for Elastic Search and Solr.
Often times real world search systems have multiple moving parts and pipelines to support query understanding, spell correction, merging results from multiple indices on the application layer etc. The final search results are only seen as the output of an application API endpoint. Rekall was born out of this particular need to integrate application endpoints to measure search quality and relevance.
- Ability to integrate API sources to fetch search results from application endpoints
- Supports GET and POST API calls with custom headers
- Ability to transform the response of an API to desired form (using javascript)
- Supports custom algorithms for ranking (You can write your own algorithms in javascript)
- Ships with four default scorers: Average Precision (AP), Cumulative Gain (CG), Discounted Cumulative Gain (DCG) and Normalized Cumulative Gain (nDCG)
- This project is bootstrapped with bigbinary/wheel.
- High level architecture of Rekall here.
- What are scorers in Rekall? How do you write custom scorers? here
- How do you transform API Responses? here
(More coming soon)
Install the latest Node.js version. Make sure that npm is installed with it as well.
./bin/setup
Start the server by executing the following command.
bundle exec rails server
Visit http://localhost:3000 and login with email [email protected] and password welcome.
- Install Docker.
- If using it for the first time, run
docker-compose buildto build the images. - Run
docker-compose run --rm web rails setupto create and seed the database. - Run
docker-compose upto start the application and get things up and running. - From now onwards, we can just run
docker-compose upfrom within the root of this directory to bring up the application.