Algolia releases new AI search features | Hacker News Reader

Algolia releases new AI search features

algolia.com

56 pointsdevin_b17 comments

I hate to be that guy, but if you’re looking for an open source alternative to Algolia, Typesense is another alternative to consider: https://github.com/typesense/typesense

It doesn’t have fancy “AI” features, but has all the other fundamentals for fast and performant search covered, including UI components.

You could also look at Sonic - https://github.com/valeriansaliou/sonic. This is used at Crisp, the customer messaging platform.

Meilisearch (https://github.com/meilisearch/MeiliSearch) is another fast open source alternative, written in rust.

Since this is indeed just a marketing page, I don't see enough implementation details to tell, but in general I worry that this kind of thing makes it easier for people to find stuff when they're being vague and harder to find stuff when they're being specific. It's like how Google these days will so "helpfully" ignore the words that you search for and instead give you what it's convinced you totally wanted.

Algolia doesn't have those vast amounts of data on the users, so they can't do too much magic.

Can I search HN, sort by date and hide all the hits which have never been shown in the top 30? Occasionally I want to find something I've seen among the titles on the HN index page some months ago.

Yes you should use https://hn.algolia.com !

I do. But how do I filter never-been-among-the-top-30 titles out while keeping the list sorted by date rather than by rating?

You can only sort by popularity, which maps to 'score' in the data structure that the API sorts. You can see that here: https://github.com/HackerNews/API The highest position on front page is not, AFAICT stored in any way and is not searchable.

In the case of HN, I assume that being once in top 30 would add a minimum number to the vote count. So a simpler query would be "sort by date and hide all the hits which have below 100 votes".

The real "have never been shown in the top 30" would need an actual event tag added to each post when they reach top 30. Then Algolia can search that I guess. Hidden tag search is a common practice.

Every search company will be expected to have machine learning capabilities. However, if you simply bolt them on top instead of considering the fast volume of additional data as part of the core engine, the experience will suffer greatly and the results will be subpar.

It was interesting to see their 30% average conversion rate improvement claim considering they lost a big customer to Sajari after we improved their conversion by 10% (and growing) https://www.sajari.com/use-cases/case-studies/catch-com-au-e...

To be fair, Algolia's marketing is great. They did a great job with the announcement. Looking forward to so how well it actually works and what the pricing is going to be. I don't think OpenAI have announced pricing for the GPT-3 service (which this is based on) yet, but I'm certain it won't be cheap to start with.

No AI yet, but hosted search & crawling at a fraction of the price: https://seekstorm.com

30x more queries & docs per dollar spent;

20x speed compared to Lucene

This seems more spam than anything else. What's new about their approach? Why is this relevant? I don't want to be mean but I don't understand why this is in Hacker News

it is on HN because it is a YC startup

It's always because people posted and they upvoted. Nothing special.

Good search is extremely hard. They provide "ok" & "fast" search very easily.

They are the official search for HN itself. That should have given you the hint. They are a YC company, so surely these posts rank up, perhaps because the alum group do some voting for peers. Which I feel is a good side-effect of being in YC.

That makes sense. Maybe that could be on the link (I did not see it) or promoted here.

Reply on news.ycombinator.com