I built a 500k-domain search engine for makers in a weekend for $10 | Hacker News Reader

I built a 500k-domain search engine for makers in a weekend for $10

alexmorleyfinch.github.io

142 pointsdreamforever86 comments

Open articleView on HN

Here's my impressions of your algorithm:

1. read each site

2. rent a 4090 with https://vast.ai to run vllm

3. let llm model invent its own category and tag names freely

4. save 1KB of metadata each

  a. a small local language model that reads each one and writes a name, two or three sentences, a category, and a handful of tags.

5. `code is going up as open source` soon (TM)

The technical details are on another page: https://alexmorleyfinch.github.io/marlin/history/v1/article/...

Your impressions seem about right, but there are a few control steps it seems.

They really needn't have specified "in a weekend" cause yeah we can tell.

Since when has low effort become a selling point anyhow?

I typically interpret it as an excuse, not a selling point.

it was by no means low effort

Isn't 60 hours by definition low effort? Setting an LLMs effort value to high doesn't count

It's an interesting new type of brag. It roughly means: "I want you to know how advanced I am at using LLMs, and how AI-first I am. So here is how little time I spent, to prove that I am using LLMs as much as possible, demonstrating that I am ahead of the curve on this new trend"

Code appears to already be up: https://github.com/alexmorleyfinch/marlin

don't forget:

6. let llm write a blog post about this conversation

Interesting project. Website discovery is indeed in a pretty dire spot, definitely a space that needs innovation. An auto-labeled website directory isn't that silly of an idea.

I have a 400 GB sqlite database with samples of rendered root document DOMs I use for ad detection in Marginalia Search I've been meaning to explore similar ideas using.

It feels like we've hit a point where search engines can become what "todo list apps" were for devs 10 years ago.

What a homebrewed solution lacks in coverage it excels in indexing and serving a small slice of the internet really really well.

To be fair they are a supremely interesting problem to hack away at, and one that will meet you where you are.

Almost anyone can put together a basic search engine in a few thousand lines of code, it's just not very hard to make a program that will index a few million documents better than Confluence.

Then, between that first ansatz and a working scalable internet search engine, you have a pile of interesting problems touching every aspect of computer science and computer hardware and networking, enough so that hundreds of people will have gotten PhDs in narrow sub-problems of those problems you'll be facing.

It's great because you can just tackle the stuff you feel comfortable approaching and leave the rest for later.

I have been wanting do do this. The biggest source of domains is certificate transparency logs. Also ICANN zone files. According to some scientific papers these cover 88% of all registered domains. You could crawl dns for CNAME records with all ipv4 IPs by distributing requests across dozens of DNS servers, the internet archive or the common crawl but doing it for the internet archive is a dick move without giving them money

There's about 200 million active domains currently. That's about 66% of all businesses worldwide of which there are around 300 million. Around 100 to 150 million have active webpages

> There's about 200 million active domains currently. That's about 66% of all businesses worldwide of which there are around 300 million. Around 100 to 150 million have active webpages

Where do you get stats on this info? Not refuting your claims but it's interesting data. I would've thought a wider gap in all biz to active websites since some countries run nearly all their biz on whatsapp/telegram/wechat.

I remember 300 million is the number of businesses on Google maps. I forgot how I got that number though. I didn't crawl Google maps. A lot or even most of those businesses run via chat often have no physical presence besides the houses of their owners so houses on Google maps are not counted. There are a lot of businesses on shopping malls and B2B businesses that aren't on Google maps often because they don't know they can add themselves to Google maps.

I did count the number of ICANN domains on their zone files on czds and there are a few services selling pre compiled domain lists that do crawling for marketers (it is forbidden by ICANN to use zone files for marketing) I don't remember how I got the active website count of 100 to 150 million, maybe I interpolated by checking some ICANN domains or there's a domain list service that does it. Domain list services usually say how many domains are active regardless of whether they have a website or not

Of course parked domains add up to the active website count even though they have no relevant content

Both IA and CCF are private foundations and accept donations.

TS;DR: Too Sloppy; Didn't Read.

Reminds me that AltaVista's servers ran in 4G of RAM (there were famous, at the time, pics of the circuit boards - DEC was rightfully proud of this, 30 years ago) and that a modern AltaVista should run on a decent laptop :-)

During AltaVista's prime, 32MB would have been a lot of memory for a typical computer.

I still can't find pics of the boards (other than on ebay :-) but https://seltzerbooks.com/alta.html has quotes like "To get a sense of the relative size of the RAM in AltaVista Search, consider that a typical personal computer today comes with 8 to 16MB of RAM, and the AltaVista Index Servers each have 6GB of RAM—about 500 times more." and "AltaVista shows the practical value of machines that have more than 4 gigabytes of physical main memory." (Dick Sites, DEC SRL) The book is from 1997.

FYI for those needing a list of domains

Subject: I want all domains and subdomains https://groups.google.com/g/common-crawl/c/XC2QmOE-sdI?pli=1

or google for COMMON CRAWL

This is actually where I see software going in the short term -- cloud moving to local.

A few years ago, if you wanted translation, you'd use Google Translate. If you wanted to search the web, you'd use Google search.

But for a few gigabytes, you can now install nllb-200-distilled-600M, and get translations for almost any language locally. You can have your computer crawl the web, create abstracts and categorizations for websites, and build search exactly as you want it.

The main limiter now is hard drive space (and to an extent, local compute) -- but right now it feels like the 70s again where the terminal into a remote server turned into building applications locally.

The number of times we've gone from cloud/server access via terminal to local compute back and forth is something that always makes me laugh a bit.

It's more of a pipeline than a back-and-forth. New abilities happen in the cloud first because they require specialized, higher capacity resources and then move towards being local as the resource usage gets optimized.

Outside insane GPU appliances, I really see cloud based things as a lock-in to subscription based offerings. The fallacy of always using the latest code isn't worth never ending subscription fees. I'll install it locally. I'll keep my data locally. I don't need to wait for things to upload/download. If there's an update that feels worthy, I'll purchased and install locally.

Sorry I have a lot of trouble understanding what this is useful for. Like, I am never going to replace it with Google, DuckDuckGo, ChatGPT or even Bing.

I was wondering the same thing. I’ve wished for just a big blob of the web to grep and regex through, but I don’t think this is that much easier than using duckduckgo or even google.

I've been working on grep.it.com. can currently grep over 200 TB of the internet

This is really neat. How much does it cost to use? I see that I can start for free, but couldn’t find anything in your docs about usage fees.

thanks. I'm still sorting out pricing. I don't have experience with B2B and am interested in empowering out-of-distribution people, so will likely maintain some free tier + congestion-based auction pricing for non-commercial use.

It's not for that, sorry, I should have been more specific. It's for people who wanna put in the effort and steer their own crawl to surface their own slice of the web. The article is just a little story of the journey

Sometimes I think people forget how capable computers are. 500k is not much. You can just slap that in a Lucene instance. This is a solved problem.

Approaching search by just tossing the data in Lucene is how you end up with Confluence's search box though.

This is a damn good project. Makes me want to make headway into an idea I've had for quite some time P2P search...we'll see.

Like a personal Google? How do you bypass all the captcha, ip bans, cloudflare turnstile antibot stuff etc?

They don't: "skips the model entirely if the page is empty, parked, or a bot-challenge wall"

Thats the fun part, the user just went with happy path. Javascript, captchas, cloudflare protected content did not made to the catalogue. This sort of use case exists in LLM training data a lot which makes it easier. The data gathered by the user is not really practically useful cause there are way too many gotchas when it comes to web scraping and building a catalogue (source: I have done scraping for a particular domain data and had to do at least 10+ iterations to get it >90 right)

despite this limitation, there is still some good stuff out there, and with the priority steering, you can focus compute on what you actually want, fast and cheap.

Check out my latest project! You can fork it, tweak the policy manually or with AI, run the system and watch the data come in! It's engineered to keep a low data footprint, so 500k domains fits into 1GB on disk. If you have local models it's free! You just might not get the best throughput depending on your GPU. My production data is not exposed anywhere yet, and I may never expose it. The point is for you to fork and make your own policy, and thus your own personal search engine! The article covers basic analysis on my data, so it's worth a read if you're interested! A deeper analysis may arrive with V2 if I ever do it

It takes a bit of setup and a huge download, but every time I need a good domain I follow this old post from Derek Sivers. I have Claude de-dupe it and turn it into a searchable database (on my machine), then have it search genres and terms I'm looking for. It's a task Claude is very well-suited to, from the technical implementation to back-and-forth about selections. [link]: https://sive.rs/com

Note -- if you do this, watch out for requesting access to "all tlds". They send you two emails per TLD -- one for your pending state, and one for your approved/rejected state. I suddenly had 1k+ emails flooding into my inbox, until I found the setting on their website to disable emails.

Holy over-engineering, Batman!

The author struggled with categorization simply because they did not truly understand k-means clustering, a fundamental concept in this kind of computing science.

You cannot just let a model run wild.

- since you know what k-means clustering is

- why dont you tell me how you ll categorize this list with k-means?

Run k-means over embeddings

> pages classified as “portfolio” or “zine” or “software” push their outbound links way up the priority list, pages classified as “corporate” or “docs” push theirs down.

And just like that, you recreated the internet of the 90s - early 00s. Brilliant!

Took me a few minutes to realise it's not a domain name search engine.

Built a free CLI for this as well: https://github.com/solozerolabs/Namera

Checks socials and trademark too

I think Kagi Small Web filter would give you very similar results.

I'll check them out!

How do you build a list of domains you want to index ? I see there is a fetcher and a spider in the code but so for I haven't found how to build that list.

Ah, I forgot to mention that anywhere. You have to provide your own. You can start from a small set, like 10 websites you like that have a bit of character, and it will also add any domains it finds from those 10

I was thinking that a search engine that ignores anything with advertising on it would be useful. This has inspired me to give it a go (on the weekend even)

Here's an idea: Each day you could summarize all the new sites into an email. Call it NCSA "What's New" or something like that. ;)

From the screenshot, it's very funny that one of the indexed sites is www.llresearch.org, which looks like it's run by a crackpot.

There are all kinds of websites in here lol. There is some gold in here and I'm determined to surface it all. I had to wrap this up without full analysis cos it was dragging on

Domains are way more than just 40M though.

From what I understand the aim was not to collect all the domains on the web but focus on personal website, etc. and avoid corporate web sites.

> There are 40-ish million domains.

citation needed

Reply on news.ycombinator.com