I love HN but always felt the search with algolia is okay but does have some limitations. Since I work at Vectara I decided to try and create a better search for HN. It's based on data from roughly the last 6 months of HN stories and comments.<p>Would love to hear feedback and how useful this is relative to the existing search.
Nice work. I wonder if there may be a better application for the Vectara capabilities than search?<p>Algolia has already done the search thing, can the Vectara search be 10x better?<p>What I do find missing from HN is the ability for me to see things that may be of interest to me, but that I may have missed. I like how I get everything in the main feed which is pure popularity, but I don't have the time to go through all posts, and definitely likely miss things I would probably have been interested in.<p>Though this can be done with collaborative filtering, or other non-AI methods, might this be a decent use case for your AI?
Like some other comments here I find HN search useful and powerful and am a little unsure what the added value is here. Possibly/probably it's for people that search in a different way to me.<p>One of the most frequent searches I do is to look for a specific comment that I know a user made recently. For example, I might want to look for my own comment here: <a href="https://news.ycombinator.com/item?id=40801389">https://news.ycombinator.com/item?id=40801389</a> (sorry, this is a slightly political one but I just picked it randomly for test purposes).<p>Searching Vectara for "n4r9 NHS" produces no results: <a href="https://hackernews.demo.vectara.com/?query=n4r9+NHS&filter=" rel="nofollow">https://hackernews.demo.vectara.com/?query=n4r9+NHS&filter=</a><p>HN's own search however produces the goods in the top result: <a href="https://hn.algolia.com/?dateRange=all&page=0&prefix=false&query=n4r9%20NHS&sort=byDate&type=comment" rel="nofollow">https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...</a><p>[ EDIT except for this very post :p ]<p>Maybe 6 days ago is outside the dataset that this is based on?<p>Some other thoughts/suggestions:<p>- Ability to click through to the comment itself? At the moment it looks like the link goes just to the main comments page and then I have to find the relevant comment on the page.<p>- Filter comments vs posts?<p>- Order by datetime?<p>- Filter within a date range?
Please, add also the possibility to search in links posted/commented in hacker news. I bet it would be competitive against google for the hn crowd.
OP mentions Algolia having limitations but this seems more limited?<p>It doesn't seem like it has any filtering or sorting like the Algolia one has, like comments/stories by a specific user, during certain dates, sorting by upvotes/recency, searching by just title/content/comments.<p>Say I wanted to search for comments by the OP, ofermend, it doesn't seem like I can...<p>Entering just their name returns results that aren't made by them nor mention their username, I tried other queries too without any luck.
Congratulations!<p>Although, something I value a lot from algolia is the very fast live search as you type[0].<p>Vectara seems to be smarter, but much slower.<p>My needs are satisfied with algolia 99% of the time as a technical user.<p>[0]: <a href="https://hn.algolia.com/?dateRange=all&page=0&prefix=true&query=hello&sort=byPopularity&type=story" rel="nofollow">https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...</a>
Thank you everyone for the feedback.
For those interested in asking questions about articles, here's another nice demo (still in beta): an Agentic RAG chatbot demo (hosted on Huggingface, using streamlit for UI):<p><a href="https://huggingface.co/spaces/vectara/hacker-news-chat" rel="nofollow">https://huggingface.co/spaces/vectara/hacker-news-chat</a><p>Feel free to ask it some things and let me know how it works.
That's interesting to hear you've developed a new search interface for Hacker News. I can relate, as I've worked on another search interface as a personal project myself (validated in "Cool Apps" at algolia), though mine has no ambitions beyond personal exploration. Knowing there are various other options already in existence, I'm genuinely curious to see how your approach distinguishes itself. Finally site hn.algolia.com is no longer maintained. <a href="https://hnpaper-labs.gaidot.net" rel="nofollow">https://hnpaper-labs.gaidot.net</a>
I searched for "Supabase" and none of top the results in your demo contained an actual post about supabase. Following the example queries I then tried "What is supabase?" and the results were equally irrelevant.<p>My personal opinion is that I'll keep using the HN search for the foreseeable time.
Cool project, but I'm struggling to understand what is better about the Vectara solution?<p>Compared to Algolia.hn, this gives 0 filter controls (time window, stories vs. comments, `author:metadat', sort order, and so on), and no ability to search for exact matches. It failed to turn up anything interesting or even relevant for the 4 or 5 queries I ran.<p>You've still made it further than I in the HN search engine adventures, which is commendable.
Cool, I like it!<p>I found a bug. Under the "When will GPT-5 be released?" search results, there are double duplicate results. On one of the duplicates, the "username (date)" says "undefined (undefined)"
I tried searching for lootitooti, Vectara found nothing, Algolia found two results, one in a post and one in a comment in another post.<p>PS: no, lootitooti is not my project. I decided to finally watch Game of Thrones with my wife and I remembered that site when I was watching the opening. I remembered seeing it here on HN, searched and found it.
It doesn't always work correctly. For example "Text to diagram tool" is returning very few results and some of the results are not even correct. While this topic has been discussed a lot here. I was mainly looking for the list of tools I keep sharing whenever this topic comes up, or when I share a related tool in a thread.
Great work!<p>I am currently playing with the Algolia hackernews search API myself and
experimenting with spaCy Named Entity Recognition and llama3 to come up with some interesting data.<p>Work in progress version here: <a href="https://news.facts.dev/topic" rel="nofollow">https://news.facts.dev/topic</a>
really like this one from [Show HN: Hacker Search – A semantic search engine for Hacker News](<a href="https://news.ycombinator.com/item?id=40238509">https://news.ycombinator.com/item?id=40238509</a>). The URL is: <a href="https://hackersearch.net/ask" rel="nofollow">https://hackersearch.net/ask</a>
Great! My first thought was "funniest XKCD". Loved the Self Driving one [0]. Thanks!<p>[0]: <a href="https://xkcd.com/1897/" rel="nofollow">https://xkcd.com/1897/</a>