I built software to read one book.
I gave up on Designing Data-Intensive Applications three times.
The third time, I built software to finish it. Six months later, that software is TextStack — open source, AGPL-3.0, free at textstack.app.
The friction
The problem wasn't the math. It was vocabulary.
Page 256 of DDIA uses "phantom" as a database isolation anomaly. The dictionary tells me it's a ghost. Google tells me it's a Rolls-Royce model. Kindle's Word Wise — same.
Every chapter has 5-10 words like that. Each lookup breaks the thread. I gave up on chapter 7 three times.
What I built
A reader that knows what book it's reading. Tap any word, get a 2-3 sentence explanation in the book's domain. "Phantom" in DDIA returns the database meaning, not the ghost.
The rest: upload your own EPUB/PDF/FB2 files; a vocabulary SRS with 5 stages (Recognition to Recall to Context to Mastered); Edge TTS audio with no API key needed; translation via OpenAI, dictionary, and full-text search.
Stack: .NET 10 + PostgreSQL backend, React + React Native frontend. Self-host with docker compose up, or try at textstack.app without signup.
Three weeks of clean data
25 unique users. 19 new, 9 returning. 32 minutes average engagement time per user. 8.2 sessions per active user. 44 Google clicks in 3 months on the broader trajectory.
Most of those 25 are people I told directly. The 9 organic strangers are scattered: US, Ireland, Pakistan, Colombia. Tiny audience, but the engagement says the ones who find it actually read.
The hard part: my audience — non-native English speakers reading technical books in English — is real but globally distributed. They're not concentrated in one subreddit or one country.
Two questions
-
If you've ever quit a technical book — what was the friction? Was it vocabulary like me, or something else I'm missing?
-
How did you find your first 100 real users when your audience isn't in one place? Open to anything that worked.
github.com/mrviduus/textstack — happy to dig into any technical decisions in the comments.
This is a strong insight. The interesting part to me is that the language barrier here is not just a writing problem, it is a meaning-preservation problem. A lot of people can explain the idea in their head, but lose confidence, specificity, or personality when they have to communicate in a non-native language. Curious whether users talk more about better wording, or about feeling like their original intent survives the translation better.
Good distinction — though for TextStack it's actually the reverse. You're describing the output side: getting your own intent into a non-native language. TextStack is the input side — reading someone else's text without losing what they meant.
Same core problem though. I can look up "phantom" and still miss the exact database anomaly the author meant. A dictionary gives me the word, not "what it means here, in this paragraph."
So honestly users don't ask for better translations — they ask for the meaning to actually land in context.
What's speechara? Sounds like you're building the output side of this.