Talksearch.io - Advanced Bitcointalk Search Engine

[NotATether]

New content update v1.0.3 and backend update v1.0.2 published

These updates add advanced search capability to Talksearch.

Search features:

• Searching by user ID, topic, board, and date is now possible.

App features:

• Added dark mode theme.
• Add advanced search section to the front page.
• Advanced search parameters appear in the query string with colons. For example: "bitcointalk author_uid:35".

48 million posts have been index now. We are approaching indexing completion.

Have you ever thought about a post/topic author rating system?

A higher ranked user - more posts, merit, ranking - has passes the filters. The rest have to go through tighter filters. This may help reduce the number of posts analyzed, and help filter better.

I don't like such a system because it will bias search results for users with a lot of merit.

[cygan]

New content update v1.0.2 published
✂️

very nice to see another update of your search engine
to update the translated threads from (taufik123, satscraper, Abdulzuruku01, katanic97, Adiljutt156, mela65, r_victory, GazetaBitcoin, Danica22 and Porfirii) i would ask you to update your op and the changelog - but you were probably planning to do that anyway

[taufik123]

New content update v1.0.2 published

Shouldn't this be a v1.0.3 update, because there was already a v1.0.2 update

New app update v1.0.2 published

[nutildah]

I think they get the overall sentiment, especially the last one, but it would be unwise to rely only on a LLM as a universal quality score.

Agreed -- what is interesting or relevant to a LLM might not be so for people actually utilizing your search engine.

Additional measures must be taken in place to identify e.g. application posts, obviously AI-generated posts, and such in order to not return them in search results.

I like the initiative you're talking here. Whats interesting is that, last I checked, Google doesn't filter out AI-generated content, but it may do so in the future if it turns out that nobody wants to read such content, thereby making their search results not as accurate or relevant to the query as could potentially be. Seems like it would be super easy to game SEO ranking with AI content, so I don't know why they wouldn't attempt to block it.

[NotATether]

New content update v1.0.2 published

Shouldn't this be a v1.0.3 update, because there was already a v1.0.2 update

New app update v1.0.2 published

You're right - But only the frontend would be v1.0.3, because there was no v1.0.2 update for the backend.

I like the initiative you're talking here. Whats interesting is that, last I checked, Google doesn't filter out AI-generated content, but it may do so in the future if it turns out that nobody wants to read such content, thereby making their search results not as accurate or relevant to the query as could potentially be. Seems like it would be super easy to game SEO ranking with AI content, so I don't know why they wouldn't attempt to block it.

I am fortunate that my preliminary tests can detect AI to a similar degree of accuracy to other types of spam.


New content update v1.0.4 published

This is a minor update that adds missing time controls for the Date From and Date To filters.

All posts up to March 2025 have now been indexed. I am actively working on enabling real-time indexing.

[NotATether]

All posts from March - June 2025 are now being uploaded to the index, while I continue to contrive an automated solution to this problem.

Edit: this batch was uploaded with wrong dates which will cause search errors, and has been deleted and is being reuploaded again.

Edit 2: All done.

I want to implement a spam score as soon as possible, but I'm still not exactly sure how I will do that without re-indexing all the posts. At any rate, I will figure something out.

[Rashlyowl]

Hey bros @NotATether, is it possible to implement pagination/paging directly on the site?



When I've gone too far, I want to go back to the page I want, but opening previous pages is a barrier for me. The solution is actually easy, just by changing:

Current page

Code:

https://talksearch.io/search?q=Bitcointalk&page=9

To

Page I want to see

Code:

https://talksearch.io/search?q=Bitcointalk&page=4

But it makes me a bit annoyed, after all, pagination can improve user experience to a better level.

[NotATether]

Hey bros @NotATether, is it possible to implement pagination/paging directly on the site?



When I've gone too far, I want to go back to the page I want, but opening previous pages is a barrier for me. The solution is actually easy, just by changing:

Current page

Code:

https://talksearch.io/search?q=Bitcointalk&page=9

To

Page I want to see

Code:

https://talksearch.io/search?q=Bitcointalk&page=4

But it makes me a bit annoyed, after all, pagination can improve user experience to a better level.

As you said, this sort of change is very easy to do, and I will make sure to find some time with it.

Due to a lack of software available on Github for this purpose, I'm currently busy building a project for calculating "embeddings" in text classification LLMs. It is a software that is blatantly missing from open-source repositories, and essential for anybody who is building a search application without a paid-for Elasticseearch subscription, which are expensive, even though they package AI search directly.

The hope is that some others in the AI community will find it useful. So I am not indexing new posts for now as the spam score calculation depends on this. Or maybe I will do that in parallel as it will benefit all. Technically, i can do it from a terminal using python scripts, but who likes doing it that way?

Edit: This is what it's going to look like: https://bert-embedding-playground.lovable.app/ - it's designed to be self-hosted. This is just the frontend though, I haven't written much of the backend yet. And even the frontend was made by AI, because I suck at designing HTML by hand.


(Modified again to avoid double-posting)

New app update v1.0.5 published

This is a minor update that adds detailed pagination to search results.

[NotATether]

In the near future, after I set up the scraper to run automatically on the index, I will set up an API for retrieving posts, topics and users and enable limited access to it.

The Elasticsearch cluster is healthy at the moment and is running without issues.

Thank you for using Talksearch.


PSA: Date filtering older than September, 2020 posts does not work properly. I will create a new index to fix this.


New content update v1.0.3 published

This is a minor update that implements a workaround to the date filtering bug described above. A long-term re-indexing will be done at a future time.

[NotATether]

Maintenance Alert

I have just received notice from the upstream provider that there might be some network-related maintenance on my node between 2025-06-28 11:00 UTC and 2025-06-29 07:00 UTC.

While I am not planning any modifications to Talksearch as this time, please be informed that access to the Elasticsearch server might be impacted during this time, and as a result, searches might fail.

The website itself is unaffected.

I will keep you guys updated in the coming days.


Edit: It seems that it was a false alarm. Nothing was affected.

[NotATether]

Gentlemen, I am pleased to announce that I have designed a local AI detection capability.



This prototype ML inferencing server which I have developed over the past week is able to download any model from Hugging Face and use it for the purpose of classifying content.

As a side effect, it can also generate word embeddings, enabling full natural-text search.

I will be performing a few optimizations on this software before I deploy it live for scoring Bitcointalk posts.

[taufik123]

Gentlemen, I am pleased to announce that I have designed a local AI detection capability.
-snip-

An amazing update, Talksearch will be updated with AI that will classify the content thoroughly, so that every content that is the result of AI can be detected easily,
this will be a very useful feature especially for project managers, it can detect which content is the result of AI and then combine it with the time span of the running campaign.

Will any content that is considered AI or some AI-generated word insertions be tagged AI as a sign that it is AI-generated?
And maybe it will be added by what percentage of AI usage.

[NotATether]

An amazing update, Talksearch will be updated with AI that will classify the content thoroughly, so that every content that is the result of AI can be detected easily,
this will be a very useful feature especially for project managers, it can detect which content is the result of AI and then combine it with the time span of the running campaign.

Thanks.

Will any content that is considered AI or some AI-generated word insertions be tagged AI as a sign that it is AI-generated?
And maybe it will be added by what percentage of AI usage.

I have no idea.

It's entirely dependent on the model I use.

There are many open-source models for AI detection, and each one makes slightly different results.

Regarding the percentage of AI, you can see it at the bottom of the screenshot.

[TryNinja]

Gentlemen, I am pleased to announce that I have designed a local AI detection capability.

This prototype ML inferencing server which I have developed over the past week is able to download any model from Hugging Face and use it for the purpose of classifying content.

As a side effect, it can also generate word embeddings, enabling full natural-text search.

I will be performing a few optimizations on this software before I deploy it live for scoring Bitcointalk posts.

I tried that before but a lot of human made content was being tagged as 100% AI, I didn't think it was accurate enough so I just ditched the idea.

I even had a friend in university getting a zero because his work was "detected as AI" with one of those online tools. He tried that with something from his professor and... it was also detected as AI, so he got his work graded.



Might go back to the idea, but I think it's better to tag something as "potential AI" than "definitely AI", becase those models aren't that accurate at the end of the day.

[NotATether]

Might go back to the idea, but I think it's better to tag something as "potential AI" than "definitely AI", becase those models aren't that accurate at the end of the day.

This is not going to be a huge problem, because AI posts will simply be demoted and not labeled.

I guess maybe I will record the percentage in the data score though so that my results are reproducible.

[NotATether]

Your opinion matters!

Do you have suggestions or improvements for Talksearch? Anything you want to see on the platform? Reply here with your idea.

Together, we well make the #1 search engine on Bitcointalk, and find topics that you have never seen before. So for this reason, community feedback is essential.

[taufik123]

Your opinion matters!

Do you have suggestions or improvements for Talksearch? Anything you want to see on the platform? Reply here with your idea.

Together, we well make the #1 search engine on Bitcointalk, and find topics that you have never seen before. So for this reason, community feedback is essential.

A great development for Talksearch.io, This is your very optimistic project and you are devoting all your energy and mind to this project.

I just want to give you some suggestions that might be useful for input or some new features such as:

Add talksearch links in the BPIP browser extension such as ninjastic.space and loyce.club that will appear next to the user's account,
which will automatically redirect to the user's post history.




and secondly, maybe you could add a Tab for the Historical thread, because as the title of the tagline talksearch.io "Explore BitcoinTalk History"
can be easily seen on the talksearch home page and several other important threads.

Maybe some kind of tab like this.

[Little Mouse]

Your opinion matters!

Do you have suggestions or improvements for Talksearch? Anything you want to see on the platform? Reply here with your idea.

Together, we well make the #1 search engine on Bitcointalk, and find topics that you have never seen before. So for this reason, community feedback is essential.

I have tried to find out topics created by a user in a specific board. I have found a way, but it doesn't allow me to filter by a date range. I think this would be a great edition.

[NotATether]

Not very important, but I was running an encoding job on the server, which kept lagging on my desktop, and apparently it can receive several gigabits of traffic as well!

Just wanted to share that with you. I thought it was really cool.

[NotATether]

New topics up to August 2025 have finally been indexed.

I apologize for the delay. This took way longer than it should have. I was dealing with some personal stuff. But we are getting closer to fully automated indexing!

These past few months have taught me how difficult it is to run a search engine, though. Much respect to anyone who can pull this off.