4 ms·
So first off personal bias, whize.co is the search engine I and my co-founder have been working on, it was nice to see basically our intended business model ent
by Grimm1 6y ago
So first off personal bias, whize.co is the search engine I and my co-founder have been working on, it was nice to see basically our intended business model entirely validated by this article. Aside from the work we've done to validate it of course.
We entirely agree with their stated aims here but believe we can produce more quality search results than many current engines, hopefully it works out.
- mdaniel 6y ago> This alpha is limited to GitHub results. ok, so: https://alpha.whize.co/search?q=KubeConfigLoader&page=1&n=10 https://alpha.whize.co/search?q=KubeConfigLoader&page=1&n=10 cheerfully says: "Showing 0 - 0 out of 0 for: KubeConfigLoader" although in your defense, it was blazingly fast to return that result. The shape of your XHR response seems to indicate you have an ES instance with only one shard, so maybe the site is still running in demo mode or something? contrast that with: https://duckduckgo.com/?q=site%3Agithub.com+KubeConfigLoader https://duckduckgo.com/?q=site%3Agithub.com+KubeConfigLoader As a meta observation, in the bottom corner of every DDG result that I've ever received is a "Send Feedback" button, allowing me to advise them of bad search results in a very low friction way, but I had to come here to report my experience with your site
- Grimm1 6y agoYup still Demo mode the index is more than a few months old at this point. We're close to launching a beta with a much larger unrestricted index and a re-crawl strategy that keeps things fresh. We've also changed our ranking methodology. The alpha / demo there was focused on novelty so if you search say "machine learning" you'd get repos that were recently active, starred and (hopefully) relevant to the topic, worth noting we also hard down-ranked repos that had been around for a while to let the smaller ones show through. Through user conversation we moved away from focusing on novelty and moved onto "information staleness" the idea that elements in search results do not decay in relevance as quickly as they should based on real world events that happen around them. Happy to talk more about it if you want.