Web and iPhone. One account. Your plan and progress follow you.
One Who Learns
onewholearns.com is a language-learning platform for serious learners, starting with German for people who need B1 for citizenship or work. It replaces the textbook-plus-three-apps combo with an AI tutor, Pip, at the center. Pip has the learner's full context, dozens of learning tools, voice conversation practice, and memory. The web and iOS apps are live with a paid Premium tier.
Highlights
Highlights
- The biggest challenge: retention Search brought a steady stream of signups. Most of them didn’t come back the next day. What the data, the screen recordings and the emails told me, and what I’m testing now. Jump to the section →
- What I had to throw away Learners found my flashcard carousel unengaging and worse for learning than Anki, where you have to recall the word before you see it. I kept the pictures and rebuilt the review around recall. Jump to the section →
- The bet: an AI tutor at the center Pip knows your level, your mistakes and your plan. It tells you what to practise next, explains what went wrong, and lets you practise speaking. The course stays for people who want to learn alone. Jump to the section →
Pip, the AI tutor
Your goal and level
What Pip sees
- Goal: Goethe B1 exam
- Level now: A2
- 30 minutes a day, 5 days a week
What he does with it
Pip says
“You’re aiming for B1, so we practise the exam tasks, not just small talk.”
Your mistakes
What Pip sees
- haben vs. sein in the past: 3 times
- Verb after ‘weil’: 2 times
- ‘die Termin’ → ‘der Termin’
What he does with it
Pip says
“Same slip as before: after ‘weil’ the verb goes to the end. Want to try again?”
Your plan
What Pip sees
- Mon: words + writing
- Thu: 10 minutes of words
- Sat: speaking with Pip
What he does with it
Pip says
“Busy Thursday? I moved speaking to Saturday. Plan saved.”
Words you know
What Pip sees
- 412 words known
- ‘Termin’: Not yet
- ‘Bescheid’: Almost
What he does with it
Pip says
“Let’s use ‘Termin’ in a sentence. It keeps slipping.”
This week’s study
What Pip sees
- Words: done
- Writing: due today
- Speaking: not started
What he does with it
Pip says
“Your words are done. Shall we do the writing task together?”
Notes about you
What Pip sees
- Works as a nurse in Hamburg
- Prefers speaking to typing
- Nervous about phone calls
What he does with it
Pip says
“Want to practise calling the clinic? I can play the receptionist.”
Problems
Problems
Apps were made for tourists, textbooks were made for paper, and everything else was scattered
Problem
Apps are made for tourists
Duolingo and apps like it teach phrases for a trip and keep you coming back with streaks. They don't take you through the grammar an exam tests.
Solution
A path from A1 to B1, with grammar and exercises for reading, listening, writing and speaking.
Problem
Textbooks are made for paper
The audio for a listening exercise was an mp3 on the publisher's website. Nothing worked on my phone.
Solution
An interactive course with the audio inside each exercise, on web and iPhone.
Problem
Plain flashcards don't use visual memory
A word and its translation on a card is rote repetition. I remember pictures better, and every German noun has a gender to remember too.
Solution
Visual flashcards where each word is a scene and the colour shows the gender.
Problem
Reference tables are scattered
Verb conjugations, articles, adjective endings: each one lived on a different website, PDF or textbook page.
Solution
Seven reference tables in one place.
Problem
Grammar drills are scattered too
Practice for one topic, like cases or articles, was spread across different sites and books.
Solution
Drills for articles, cases, verbs, word order, prepositions, pronouns, adjective endings and modal verbs.
Problem
I wanted it all in one place
Words, grammar, practice and my progress, on my laptop and on my phone.
Solution
One app on web and iPhone, plus a Chrome extension that saves words from any page.
Who it’s for
- Decision
- Build for learners with an exam to pass, starting with the B1 citizenship exam.
- Why
- They have the most urgent need and a finish line you can measure. I am this learner, so I could test the product on myself every day.
- What I left out
- Travellers and casual learners. There are more of them, but their motivation fades fast and the big apps already serve them.
- How I checked
- A customer journey map and an offer built from forum research and competitor apps in March. In September I read every answer new learners had given in onboarding to one question: why are you learning German?
What 62 new learners told me
Why are you learning German?
| Study | 100% |
|---|---|
| Work | 84% |
| No reason given | 84% |
| Citizenship or a visa | 63% |
| Family, partner or friends | 38% |
| Travel | 38% |
62 answers from new learners, June to September 2026.
Day 2 retention, by the level they aimed for
| Aiming for B1 | 100% |
|---|---|
| Aiming for A2 | 44% |
| Aiming for B2 or C1 | 28% |
Learners who used the goal picker, September 2026.
Most of the users named an exam or a level, so the niche held.
Two things surprised me.
More people learn for study and work than for citizenship. And many aim for B2, one level above what the product teaches. Learners aiming for B1 came back the next day far more often than learners aiming higher.
So I kept the product at A1 to B1 and made the ceiling honest: you can pick B2 as your goal, and the app and Pip tell you what it can and can't take you to.
Architecture
Architecture
How Pip, the AI tutor, works
It took a few months of thinking, research, design and building to work out what the learning system should be. Three versions got it there.
V1
Tools
A set of tools for learners.
- Vocabulary
- Reference tables
- Courses
- Drills
- Study plans
V2
A learning system
AI runs your study.
- AI plans what you study next
- Tools like Anki flashcards and grammar drills
- AI writes new content when the library falls short
- Planned from your learner context
- A levelled library for reading, listening, writing and speaking
V3
AI tutor
The icing on the cake.
- Pip: a tutor character and speaking companion
- One log of mistakes and progress
- More tools for Pip to manage your context and study
- Memory notes, so each chat picks up where the last one ended
Learning system v3
Claude / ChatGPT their own AI
MCP adapter same tools
Web chat
iOS chat
Pip Teaches from the context. Suggests the better step, once.
Look up read only
Make their words + mistakes
Task or session with a “why”
Card in chat
Today list
Practice dialog over the chat
Result score + misses
Progress every answer
Weekly email idea · reads the learner context
Learner context
One per learner, on the server. Loaded once per chat, then only new parts are added.
- Word memory
- Plan
- Pip’s notes
- Learning style
- Goal + deadline
- Gap to goal
SCORES FROM EVERY ANSWER
- 4 skills
- 2 blocks
- Weakest — gap to goal
Getting found
I built two ways to reach learners: search took four months to take off, and social media still hasn’t.
I funded this myself, so I chose channels that cost time instead of money. Search came first. The big apps rank for almost nothing about German grammar, vocabulary or exams, and people searching for those are already trying to learn. Social media came second, built from the same pictures as the flashcards.
Search pages
- Data
Search Console, Bing and Ahrefs keyword data
- Plan
Page types ranked by demand
- Pages
Built from the word database
- Measure
Clicks and signups feed the next plan
↺ feeds next plan
Social media
- Words
The same database and pictures
- Five formats
Word cards, phrase carousels, word sets, grammar scenes, der/die/das videos
- Review
Every post checked and posted by hand
Built for Instagram and TikTok, with a renderer for each format. Nothing posts automatically.
Visitors by source
LAST 180 DAYS · 7-DAY AVERAGE
- Organic search 345/day
- Direct 212/day
- Referral 34/day
- AI assistants 15/day
- Other 10/day
Bing moved first, in July. Google followed at the end of August, and by September the page with the 1,000 most common German words was on the first page of results. Social media started from zero followers and went out by hand when I had time, and it still sends only a few visitors a month. Search also brought a problem I didn’t expect, which is the next part of the story.
The biggest challenge
The biggest challenge
Signups grew fast,
and most people never came back the day after
Search worked, and signups climbed through August and September. So I followed what new learners did after they signed up, step by step, through the product’s event log. Most of them stopped early. Few came back the next day.
How far the first signups got
When learners came back after signing up
What I found, and what I changed
- 01
People came for a PDF
The page that brought the most visitors was a free list of the 1,000 most common German words. Its visitors signed up four times less often than visitors to the home page. They took the list and left.
WHAT I CHANGED
The app now remembers what brought you in. People who came for words can start with words, or with Pip.
- 02
The first day asked too much
One learner went between the home screen and lesson 1 six times in three minutes. Over half of new learners never saved a single answer.
WHAT I CHANGED
Setup builds your plan and ends on your first task. Home leads with your goal and one row for today.
- 03
Finishing setup went with coming back
Learners who finished setup came back the next day about three times as often as those who skipped it.
WHAT I CHANGED
Pip now runs setup as a short conversation.
- 04
Nobody heard from us
A new learner got no email after signing up. And when reminders did exist, three in four of them went to people who had already stopped using the app.
WHAT I CHANGED
Emails on days 3, 7, 14 and 30, plus a next-day email from me to anyone who skipped setup. Reminders stop after two quiet weeks.
- 05
Some people only need one job done
Not everyone wants a course. Some people just need to check a word.
WHAT I BUILT
A dictionary: look up any German word in any form and save it in one tap. Over 100,000 entries.
THE STRATEGY
Retention isn’t solved yet. I work on it as a loop.
Each round starts with real learners and ends with something new to measure.
- 01
Listen
Talk to learners and watch session replays to see where they get stuck.
- 02
Redesign
Rebuild the parts that don’t work, one at a time.
- 03 ↺
Make new bets
Try new ideas, see what brings people back, and start again from 01.
Learning from users
“I don’t want to think about what I’m missing and what I’m good at. The system should care about that, not me. My mind should be free for learning.”
Learning from users
I built the product on assumptions. Talking to users refuted some of them.
In July and August I emailed around 80 new learners and asked what they were learning for and what was missing. Some wrote back with requests. I booked calls with them.
Each shared the way they use the app and what’s missing from it. I’ve got lots of ideas and suggestions how the app can be improved further.
ASSUMPTION WHAT I LEARNED WHAT I CHANGED
ASSUMPTION A weekly plan, a daily goal and a daily email keep learners going.
WHAT I LEARNED The learners I spoke to did not use the plan or the dashboard. One said the daily email made him feel guilty about not studying.
WHAT I CHANGED Home now starts with today’s tasks and readiness answer. Daily emails are user context aware and tell the user what to focus on today, without putting a pressure.
ASSUMPTION Progress means words learned and lessons finished.
WHAT I LEARNED They wanted one answer: am I ready for the exam? One learner: “I don’t care how many words I memorize. I don’t know if I’m ready or not.” His driving test app told him when he was ready.
WHAT I CHANGED A readiness line already existed, at the bottom of the Roadmap page. He had never seen it. It is now a card on Home.
ASSUMPTION The app is the course.
WHAT I LEARNED Their course is a textbook in a classroom, or a tutor. The app is the tool beside it. One learner called my course section useless, because it didn’t follow the lesson he was on.
WHAT I CHANGED I had stopped scheduling course lessons a few days before the call. The call confirmed it. Personalized courses made by the AI tutor are on the roadmap.
ASSUMPTION A quiz should mix in new words to grow vocabulary.
WHAT I LEARNED They study the words from their current lesson and want to be quizzed on those, plus a few old ones. New words in the quiz looked like a bug.
WHAT I CHANGED First, I fixed the UX: you now choose a set of words first, then learn and quiz only that set. Later I replaced the whole Vocab workflow with one based on Anki.
What I took from it
Three of the four beliefs had the same shape: I had built something that asked learners to plan, track or choose. They had around 30 minutes a day and wanted to spend them on German. That changed the question I ask about every new screen. It used to be “what does this help you learn?”. Now it is also “what does this ask you to manage?”. It is also why the next section of this page is about a tutor and not about another feature.
Re-design
Re-design
When something doesn’t work, you either cut it or fix it, no matter how much effort you have put into it.
Flashcards have been the core of the app since the beginning. Visitors who tried them signed up far more often than visitors who downloaded a PDF: 5% against 0.9%. However, users did not come back to them. Four in ten flashcard sessions ended without a single card graded. It felt like users believed the promise, but it wasn’t fulfilled.
That’s when I started to examine the issue by looking at the usage data and replays. I discovered that the UX and UI I designed didn’t quite work. People would click through the slider and leave.
At that moment I thought: why not take a proven and popular mechanism, Anki, and improve its weaker sides by adding color coded visuals and core word information. I also added directionality, German to English and English to German, since recognising a word and producing it are two different skills.
- HYPOTHESIS
- The images work but recalling words in a slider doesn’t.
- IF TRUE
- The share of sessions with at least one graded card goes up from 60%, and learners who grade cards come back the following week more often than before.
- IF FALSE
- Grading stays where it was. Then the problem is in the words or the pictures, and the next change is to the content.
- WHEN I’LL KNOW
- After 500 sessions on the new cards, around mid-November 2026.
The bet
The biggest bet: an AI tutor that knows your whole learning context
The best part of my own learning was my tutor. She looked at where I was and picked the exercises and grammar drills that fit. Whatever your goal, there is a fixed set of skills you need for it. A tutor who knows which ones you have and which you don’t can replace a fixed course. I built that tutor into the app and called it Pip.
It is the most expensive part of the product to run: every conversation costs model and voice minutes. And it has no usage data yet. That is what makes it a bet.
ALONE
You don’t know what to do next
WITH PIP
Plans your week from your goal and weak spots, and tells you the next step.
ALONE
You don’t know what’s wrong
WITH PIP
Reads every answer you’ve saved and finds the gap.
ALONE
Nobody explains it
WITH PIP
Teaches by talking, then checks that you understood.
ALONE
Nobody notices
WITH PIP
Remembers what you struggle with, and reacts when you slip or improve. You can see and edit what it remembers.
How Pip teaches
Explain another way
- 01 Know you goal · gap · memory · habits
- 02 Find the gap what blocks the goal now
- 03 Teach — talk explains, answers, lesson + word cards
- 04 Check a quick question in the chat Understood: on to practice. Not yet: explain another way, then back to teaching.
- 05 Practice drill · task · session
- 06 React sees the result, responds at once
- 07 Remember the context updates
First principles
- 01 The AI tutor knows you, and shows it Every suggestion names something true: the goal, the gap, a past mistake.
- 02 The conversation is the classroom Concepts are learned by talking, like with a real tutor.
- 03 Practice at the right moment After understanding, when asked, or when the goal needs it.
- 04 Never break the thread Practice opens over the chat and closes back into it.
- 05 The learner leads, the tutor guides Does what is asked; offers the better step once; asks before changing the plan.
- 06 Everything teaches Pip something Chat answers, drills, readings all update the context.
- 07 Pip learns how you like to learn Talk or drills? Watches which offers are taken.
- 08 One brain, many doors Web, iPhone, Claude / ChatGPT share context and tools.
- 09 A tutor is three parts Context + tools + instructions. German now; other subjects later.
The decisions behind it
Context is the product
Without your history, Pip is a chatbot. Everything Pip says starts from your goal, your answers and your mistakes.
Exercises open inside the chat
Pip starts in a chat, and exercises open inside the conversation. You never get sent somewhere else and lose your place.
Reuse before making new
Pip uses the lessons, words and exercises that already exist before it writes anything new.
Do what’s asked, suggest once
If you ask for something, Pip does it. If there’s a better next step, it says so one time.
Home works without AI
Pip is the home screen, but the home screen loads without waiting for a model. The library stays for people who want to learn alone.
Voice as separate steps
Talking to Pip runs as speech to text, then Pip, then text to speech. Voice minutes and credits have limits, and misuse gets warnings before a block.
- HYPOTHESIS
- Learners who start their day in the chat with Pip come back more often than learners who start on the library.
- IF TRUE
- More Pip users come back in their second week than library users. And they start more practice from the chat than from the library.
- IF FALSE
- Pip is a nice chat, but not a reason to come back. Then the next step is back to the retention loop, and Pip stays a helper inside it.
- WHEN I’LL KNOW
- After 50 learners have had Pip for four weeks, around December 2026.
What’s next
Next, I want more learners to stay and find the app invaluable in their learning journey.
- 01
Solve retention
Measure how many learners come back the next day and the next week, for the new setup, emails and flashcards.
- 02
Polish
Release the new flashcards and Pip together, as one update people can notice.
- 03
Talk to learners
More conversations like the one that led to import and Android. What do people actually use OWL for?
- 04
Then grow
Android on the Play Store. Pages for people learning for study and work, the two largest groups.
The product today
Everything above adds up to this.
A structured course, A1 to B1. Grammar explained in plain words, then exercises after every rule.
Add any word. AI fills in the translation, a picture and an example sentence, and the word joins your reviews.
Save words from any website. The Chrome extension translates as you read & sends new words to your deck.
Every grammar table in one place. Articles, cases and verb forms, plus unlimited drills to make them stick.
Bring your own AI. Connect ChatGPT or Claude to your words and progress, and practise there too.
Unlimited grammar drills. Practise articles, cases, word order and verb forms until they stick.
