I often think about this precise scenario. There are so many cases and times where a big company shuts something down that would be perfectly viable at a different scale. My previous company got acquired by Verizon and then shut down because it didn't scale to a $1B business within five years (or something like that). I'm not claiming it was a viable business in the first place (it probably wasn't) but now we'll never know. And for this story, there's a thousand others. I understand that sometimes the small company becomes intertwined with the big company to a detrimental extent, but there are also plenty of cases where a separation is still very well possible. The separation doesn't even need to be the business whole, but it could be a constellation of IP and resources, for which someone creates a new business entity. I wish that more execs created that possibility, instead of just writing off a financial loss and calling it a day. While the write-off makes business sense to the bigger company, and there may not a huge business justification for spinning off instead of shutting down, there's also just something to doing the right thing in terms of providing continuity to the users of your product, and giving engineers a chance to double down on a product development that they may have faith in.

Yep, I've got nothing but good things to say about Duolingo :-)

I would never say this in the Balkans but maybe take a look at Croatian/Bosnian resources as well.

Nobody wants to admit it but it is basically the same language, heck was called Serbo-Croatian a few decades ago.

> You have produce a connection to my heritage and ancestry that nobody else has, and I really appreciate it.

Awww, this is such a precious sentiment <3 it always makes me happy to see people's lives changed like this online~

That's wonderful to hear! I hope you find it useful. If you have any feedback after using it for a while please let me know!

Thanks for the suggestion! I think in that case I'd probably want it to automatically combine certain adjacent words into phrases, which is definitely doable with AI but could be expensive to do for an entire text vs just paying for the words and phrases you click to translate like it does at the moment. Also, newsinslowspanish only turns some of the content into links - it makes an assumption as to which words you'll know and which ones you won't, which may not be valid. If you turned the entire text into their style of hoverable links it could look cluttered. You'd want more subtle ones. In fact, it would be very much like Duolingo Stories, which does work well! I'll think about this.

Good point! I just made the first mention of Readlang into a link to the website: https://readlang.com

It's a web-app and browser extension to help people learn languages by reading texts or websites in the language they are learning, get helpful translation hints and definitions to aid comprehension, and then to practice the new vocab with flashcards.

Tucan actually works in reverse compared to Readlang. Say you're a native English speaker learning Spanish. With Tucan you browse the English websites you're already used to browsing, and it'll automatically translate some of the words to Spanish to sprinkle in a bit of practice that fits within your normal routine. It's a nice idea (actually this is the first one I saw doing the idea: https://www.theverge.com/2012/5/3/2995523/google-language-im...) but for me it didn't stick. I found it weird to be learning isolated Spanish words out of their normal context, and pretty soon it just became annoying to have these Spanish words popping up everywhere. Maybe it works for some people though.

With Readlang the idea is that you read actual websites in the language you're learning. If you're learning Spanish you might read El Pais or BBC Mundo for example. And to aid you, you can click on the odd word or phrase you don't understand to get a translation in English. It's intended for intermediate and advanced learners. Unlike Tucan you need to seek out texts to read in your target language, and Readlang offers a list of popular sites in each language to help you there: https://readlang.com/es/links, but there's obviously more friction than something like Tucan which fits into your normal English language browsing routine.

The ideal fix was to upgrade the version of Stripe checkout. At the time though I was looking for a quick solution and was scared to update the Stripe libraries for fear of accidentally breaking something else. So here's what I did:

- First I tried adding an IP address based rate limiter to the signup and payment routes, and required that all users had translated at least one word before being allowed to upgrade. It stopped the attempts for a few hours, but then they started back up again.

- Next, I used a more drastic approach, which would prevent all payment attempts from all IP address if we received more than 10 failed payment attempts within a 1 minute window. Still didn't work.

- Next, I changed the thresholds to 4 failed payment attempts within 5 minutes, and this finally worked!

Since buying Readlang back, I've updated the Stripe checkout to the latest version, turned off the circuit breaker, and haven't had any further problems.

There a lot of ways to reduce, just some ideias: turn on 3dsecure authentication whenever available. Check user location vs card location. Validate cvc code and postal code. Check for user activity and user signals (user agent, screen size etc) on the website, and other methods.

Yep, I like the fact that it's a simple business model with well aligned incentives. It's a direct relationship between me and my customers. I keep getting paid as long as the site continues to be useful.

I also send an automated "Readlang account renewed" email every time I receive a renewal payment. Most SaaS products don't do this for obvious reasons, but it sucks when I realize I've been paying a subscription for something I'm not using, and I don't want to do that to my customers.

What everi

Sorry to hear that.

I imagine it would have been a lot harder in my case if we'd actually integrated Readlang with Duolingo, sharing user accounts and that kind of thing. Luckily it was quite separate. I was a little stressed dealing with accountants and tax stuff here in Spain, and worrying that the technical transfer would go smoothly. But dealing with Duolingo was super easy.

Yeah I use the Google Translate API for the translations.

You can also embed a custom dictionary of your choosing in an iframe in the sidebar: https://blog.readlang.com/2013/11/07/custom-dictionaries.htm...

And recently I've added a feature to get context-aware definitions from the ChatGPT API, but this is just for premium users in English, Spanish and French for now.

You can use the bookmarklet, which you can drag into Firefox's bookmarks bar from this page: https://readlang.com/webReader

It works almost exactly the same as the Chrome Extension.

It's mainly for intermediate and advanced learners who want to tackle longer texts. I originally made it so that I could read novels in Spanish.

Readlang may not ever be appropriate for complete beginners. But it could do better at providing stories written in simple language appropriate for beginners in specific situations like yours. Right now there is a public library of shared texts: https://readlang.com/fr/library, and a list of popular websites: https://readlang.com/fr/links. But you're largely left to your own devices to find content, which adds friction for new users and isn't ideal.

It was a mix of cash and stock options which vested over a few years. There was a confidentiality clause in the contract so I can't be super specific. But I asked ChatGPT what a typical price per employee is in an acquihire and it gives the following answer. I'll just say that the common range it gives doesn't sound crazy to me.

> In general, the price per employee in an acquihire can range from $100,000 to over $1 million, with a more common range being between $250,000 and $500,000 per employee.

Basically it's because I don't think Duolingo wanted the product so much as they wanted me. No-one used the term acquihire at the time but I was under no illusions that that's what it was.

I did think that I might push to develop Readlang at Duolingo, but it never seemed like an attractive proposition for a bunch of reasons. I think it would have needed to grow at least 100X to start getting interesting within Duolingo. Not to say this was totally impossible but it seemed like a tough job. Also, some of the things that Readlang does, like browser extension and public sharing of texts, feel a bit "wild west" compared to the rest of Duolingo which is more curated and controlled, so it's not clear exactly how Readlang would have needed to change to fit in. I figured that if I tried this and integrated it tightly within Duolingo, and then failed to grow sufficiently, then the risk of it ultimately shutting down would have increased.

Also, I had other exciting projects to work on there which seemed more valuable for Duolingo, and therefore better for my reputation within the company, so I was happy to work on those!

> if it was really a simple low ball for duolingo to get rid of

At 4K MMR it was probably a pain in the ass for Duolingo to maintain and the reason for acquisition was hiring the developer and avoiding a potential concurrent (which in the end never realized) so I'd imagine it was a very easy buy.

And I'm not saying that as a criticism of ReadLang, I think it's a great tool actually.

Yeah, this is a problem. Anyone is free to share texts to the public library and there's no quality control, just an upvoting feature to try to surface the best content at the top. I accept that the public libraries are a mess at the moment and it's not easy to identify the high quality content there. Lots of room for improvement!

Good question! There was a lot to do and I was kept busy just with internal communication, so didn't feel the urge to share externally so much. Also, getting outside attention didn't feel as important given my role at the company compared to when I was alone running my own project.

I actually remember asking them how restrictive they would be when it comes to sharing when I joined and they said they weren't as secretive as Apple, but that they weren't completely open about everything either. If I had had the urge to post anything I would have run it past people before publishing details about what our team was working on, and it would have probably been fine. But in the end, like I said, I was always busy and blogging fell by the wayside. (I was never a naturally prolific blogger anyway)

Hmmm... I'm actually not sure if I can talk about the numbers from the original sale in 2017. But I'll say that the original sale was paid partly in stock options which vested over a few years and ended up being worth a lot. And the price of the buy back was much much less than that.

Funnily enough, in the original sale I negotiated for more cash and fewer options. Luckily for me they only moved a little bit in my direction since the options ended up being worth much more.

Not a ton of methodology behind it!

I started at $10 per year, then gradually raised prices as the product got better, always keeping existing subscribers happy by grandfathering them. At one point it was $4.99 every 3 months, which is the plan my mum is still on :-) For a brief time I offered a lifetime subscription for $100 and sold a few but felt uncomfortable offering to keep the service running forever with no further payments so killed it.

I ended up at $5 / month or $48 / year. It's possible I might increase it a bit more in future but for now I think it's better to keep these prices and try to grow user numbers instead.

One thing I noticed with some companies was that the yearly plan was very heavily discounted compared to the monthly one. To me that gave the impression of lacking confidence that monthly users will stick around long term, so I didn't want to do that.

Yeah I think it's true that popular mainstream consumer products tend to be cheap and highly polished, and that more niche products tend to be higher priced and less polished.

For now it provides the whole sentence that the word was in for context. It works pretty well but I'm sure it would be better if I gave it a little more. 2 or 3 sentences would probably be great, but obviously more expensive to run.

Getting close to 5k yeah. The average over the preceeding year was around 4k.

It's very verbose by default, but that's when/why you use it. If you just want to know why something doesn't work, you use ping.

    $ ping nxdomain.example.com
    ping: nxdomain.example.com: Name or service not known

    $ ping example.com
    PING example.com(2606:2800:220:1:248:1893:25c8:1946 (2606:2800:220:1:248:1893:25c8:1946)) 56 data bytes
    64 bytes from 2606:2800:220:1:248:1893:25c8:1946 (2606:2800:220:1:248:1893:25c8:1946): icmp_seq=1 ttl=52 time=93.2 ms

If you want a short summary, you use host:

    $ host nxdomain.example.com
    Host nxdomain.example.com not found: 3(NXDOMAIN)


    $ host example.com
    example.com has address 93.184.216.34
    example.com has IPv6 address 2606:2800:220:1:248:1893:25c8:1946
    example.com mail is handled by 0 .

or nslookup (from the same bind9-dnsutils package as dig):

    $ nslookup nxdomain.example.com
    Server:  127.0.0.53
    Address: 127.0.0.53#53
    
    ** server can't find nxdomain.example.com: NXDOMAIN

    $ nslookup example.com
    Server:  127.0.0.53
    Address: 127.0.0.53#53
    
    Non-authoritative answer:
    Name: example.com
    Address: 93.184.216.34
    Name: example.com
    Address: 2606:2800:220:1:248:1893:25c8:1946

If you want to debug DNS, with all the raw data, then you use dig.

Alternatively, use dog (https://github.com/ogham/dog)

> dog A readlang.com

A readlang.com. 1h57m12s 139.144.234.197

> dog A someinvalidserver.com

Status: NXDomain

Yes, it's horrible. It helps if you know the actual syntax used in zone files, and yes, it's the lack of an answer section.

Alternatively use the short format as suggested by another comment:

    $ dig a www.readlang.com +short
    readlang.com.
    139.144.234.197

Although now it's a bit too short and you can't see the record types (CNAME, A). Sigh.

+short is your friend.

Out of curiosity I asked ChatGPT what the output means and this is what it responded :) Maybe this helps you!

This output is the result of using the dig command to query the DNS (Domain Name System) for an A (Address) record associated with the domain name "www.readlang.com". Here's a breakdown of the output:

; <<>> DiG 9.10.6 <<>> a www.readlang.com: This line shows the version of dig being used (9.10.6) and the command executed (dig a www.readlang.com). The "a" argument is asking for the IPv4 address associated with the domain name.

;; global options: +cmd: This line displays global options for the dig command. "+cmd" is a default option that includes the initial command line in the output.

;; Got answer:: This line indicates that the DNS server has provided an answer to the query.

;; ->>HEADER<<- opcode: QUERY, status: NXDOMAIN, id: 65131: This line shows details about the response header. "opcode: QUERY" means it's a standard query, "status: NXDOMAIN" indicates that the domain name does not exist (non-existent domain), and "id: 65131" is a unique identifier for this query.

;; flags: qr rd ra; QUERY: 1, ANSWER: 0, AUTHORITY: 0, ADDITIONAL: 1: This line provides information about the response flags and the number of entries in each section of the response. "qr" means the response is a query response, "rd" indicates recursion was desired, and "ra" means recursion is available. The number of entries in the QUERY, ANSWER, AUTHORITY, and ADDITIONAL sections are 1, 0, 0, and 1, respectively.

;; OPT PSEUDOSECTION: and ; EDNS: version: 0, flags:; udp: 1280: This part shows information about the Extension Mechanisms for DNS (EDNS). The version is 0, and the UDP payload size is 1280 bytes.

;; QUESTION SECTION: and ;www.readlang.com. IN A: This part displays the question section of the query, which is asking for the A record (IPv4 address) associated with "www.readlang.com".

;; Query time: 62 msec: This line shows that the query took 62 milliseconds to complete.

;; SERVER: 2406:3400:41e:5a80::1#53(2406:3400:41e:5a80::1): This line displays the IPv6 address and port number of the DNS server that responded to the query.

;; WHEN: Fri Mar 24 10:06:36 AEST 2023: This line indicates the date and time when the query was executed.

;; MSG SIZE rcvd: 45: This line shows that the total size of the message received was 45 bytes.

The main takeaway from this output is that the domain name "www.readlang.com" does not have an associated IPv4 address (A record) in the DNS, as indicated by the "status: NXDOMAIN" and "ANSWER: 0" parts.