Major_Grooves

meanwhile you practically have to pay people to take away current used Ikea pieces.

The market for old DDR (East Germany) furniture here in Berlin is a bit crazy. Old cheap poor quality furniture (all MDF) selling for loads because it looks good on Instagram.

I know a guy that goes to house clearance sales in random places in East Germany, then brings pieces back to sell to hipsters in Berlin. Like me.

I'm the founder of Tilores, the entity-resolution tool used here - so full disclosure, this is my company's product. This wasn't a paid engagement or a case study. It started because my wife is from Venezuela, and she saw people on social media pointing out that the missing-persons lists had huge numbers of duplicates.

On the data: these are public citizen-lead efforts to crowd-source the names of the missing - hosted on websites and spreadsheets. There is no official verification process behind the individual entries, which is part of why the duplicate problem existed in the first place.

An issue we have now realised is "bad actors" trying to access the data...

Happy to answer anything - methodology, false positives, data handling, whatever.

I am only slightly miffed that you did not mention your other co-worker who, before the co-worker you did mention, but an entire two-sided marketplace using LLMs (lovable). https://minnnis.com/

Please make it work with Lovable for vibe coders like myself.

Do you end up with lots of duplicates when you are scraping? If you also scrape IG, YouTube and LinkedIn, would you link them all to the same influencer?

That might be quite an interesting identity resolution challenge (disclosure: I build identity resolution tech).

I would not mind taking a look. Always interested to see how others are handling such data.

Hello HN! We (Steven, Hendrik and Stefan) built a real-time identity resolution system that can handle hundreds of millions of customer records, and recently launched a LangChain integration to use it as a RAG source for LLMs.

We built this while working at a European credit bureau, where we needed to deduplicate and match millions of monthly record updates from various sources. Traditional approaches using graph databases and Spark couldn't handle the scale, so we built our own solution using AWS Serverless.

Each identity is stored as an individual graph structure, using rules-based and ML matching. Performance: <300ms ingest (tested to 5,000/sec), <150ms search regardless of graph size. Several fintech companies use it for fraud detection, KYC, and customer 360.

Unlike vector databases which can blur similar entities together, IdentityRAG maintains distinct customer identities while pulling data from multiple systems - even when customer details differ across databases.

You can try it out with our sample chatbot in the Github repo (linked above). Free to sign up, we charge based on number of unified customer records (it is free for playing and testing). We would love to hear your comments and questions.

There is also a demo video in the repo and you can find more details about us here: https://tilores.io/

I experienced the same thing, and wrote nearly the same article two years ago: https://major-grooves.medium.com/just-how-complicated-could-...

Yes I know we can register a UG, but in the end you don't. And it is not just the share capital that is annoying it is everything else.

It literally costs 10x more to do the bookkeeping for a German company vs a UK one. Plus getting investors is much more difficult because of the notary requirements.

We used a SPV for our first round, but even that is annoying. I had some angel investors pull out purely because we are a German GmbH.

https://tilores.io/ is our original solution

https://tilosearch.com/ is when prompted people to ask us about RAG

I have no idea what this is.

Sounds a bit like Blinkist for the classics. Nice idea.

just to be clear - the 61 duplicate voting cases were only for Ohio and Pennsylvania - the 400k duplicate profiles were across all 7 states we looked at.

Indeed there is certainly not mass voter fraud. We were glad not to find that, but tbh surprised that we found any at all. Originally we were only going to look for duplicate profiles - it didn't even occur to us to look for actual fraud.

But why not make it a complete non-issue? It would be so easy to fix this data so there were no duplicates, then there would not even be any accusations like there were in 2020.

What I want to create is complete trust in the data to avoid the... bickering later.

/edit - as the poster below mentions, the 61 were just the ones that were manually confirmed. There were 1000 potential cases.

We come from Germany - where there is unlikely to be a big issue, as citizens have to be quite careful about registering where they live in one place only.

I suspect the data in the UK (where I originate) would be pretty messy. The voting lists there are a free for all, I reckon!

This is a problem I also have as a founder of a German company. Meetup literally can't add the company name to the invoice unless it is in the credit card name.

Ok this could actually be quite handy.

Indeed, unfortunately with the John Smiths of this world there will be false positive matches. What we could do is add that they need to be from the same town/postcode, but then that is quite an unreliable attribute too.

Similarly, there are a lot of false negatives where we know two records should match, but we could not because that would require a rule that would create more false positives.

In the end, it was the best we could do with the public view of the data. If we were working with the data Companies House actually holds itself, it would of course be much better.

My family business in Scotland is in the construction sector. I will pass this on to them.

This is really great. I am a big fan of Lottie files and a big fan of anything that makes them easier to work with.

I am so glad to hear about this. Readlang has always been one of my favourite language tools and I am sure you can grow it up to to be a pretty decent sized service, with decent MRR.

Let me know if you are ever in Berlin and we can meet up again!

I think we prefer "restrained" to "meek" ;)

just to be clear - we did not expect to catch him, get him out of the car, and make a citizen's arrest. We just wanted to know what direction he went in to tell the police, but he was too fast.

it's Scotland. Making a citizen's arrest is not so crazy.

in this case I really meant that the act of deleting and reinstalling the app is the fraudulent act, although it is not what most people would consider fraud (i.e. financial fraud) it would normally fall under "trust and safety" I guess.

Thank you for your sensible reply.

exactly - under GDPR it would be fine to retain specific data for a limited period of time to prevent fraud (which is what this is, really).

Also, this type of user is not making deletion requests - they are deleting and then immediately reinstalling the app to appear "fresh".

thank you for the insight.

Indeed the same though crossed our minds (as I write in the article) that it is unlikely to be his first offence.

We actually did not want to humiliate him, and believe me we had the opportunity, because we did not want to "push him over the edge". I kinda hope he has kept his job so he can try to get help and stop himself doing this again.

aye, exactly. I presume most of the people criticising our "citizen's arrest" idea are form the US, where I indeed would not have contemplated doing such a thing.

In Scotland I do not really have the same fear. Not sure I would do the same thing in Glasgow or London, but still, we are not idiots, we were not going to take any crazy risks but we figured the risk to us was worth it to stop the risk to our sister. We weighed the risks up.

"Complete idiots"might be a slight exaggeration, but hey no offence taken.

iirc they knew we had video, but the video of the car vandalism was never going to be enough to do anything. They had not seen that specific video. In the end it was only a small part of the evidence package.

Not necessarily. If it is for fraud prevention, then you would be within your rights (at least under GDPR), to retain certain data for a reasonable amount of time.

Most people indeed commented that we showed quite decent restraint.