78 cards, 20 spreads, 5 languages: the Tarotas by Inithouse build breakdown

We run a product studio called Inithouse. One of our products is Tarotas, a tarot reflection app where you draw a card and read a grounded interpretation. No signup, no fortune-telling, no paywalls. Just space to think.

This post breaks down what "78 cards in 5 languages" actually meant to build, the content decisions that shaped the product, and what we learned about daily rituals and retention along the way.

The content math nobody warns you about

A standard tarot deck has 78 cards. Each card has an upright and reversed meaning. We wrote a grounded, psychology-informed interpretation for each. That is 156 pieces of content before we even started thinking about languages.

Then we added five languages: Czech, English, Polish, Slovak, and German. That multiplied the interpretation set to 780 individual texts. But the real number is higher because Tarotas also supports 20 different spread layouts, each with positional context that changes how a card reads. A card in the "past" position reads differently than the same card in "advice."

Here is how the content stacked up:

| Content layer | Count | Notes |
|---|---|---|
| Card interpretations (upright + reversed) | 156 | Base set, ~150 words each |
| Language variants | 780 | 156 x 5 languages |
| Spread layouts | 20 | From single-card draws to 10-card Celtic Cross |
| Positional context modifiers | ~200 | Per-position phrasing across spreads |
| Daily affirmations | 78 | One per card, rotated daily |
| Ask-tarot (yes/no) | 78 | Single-card draws with binary framing |

Total addressable content surface: north of 1,100 individual pieces. For a product that looks like "flip a card, read a paragraph," the backend writing effort was substantial.

Translation was the first real wrong turn

Our initial approach to the five-language set was straightforward: write everything in Czech, translate to English, then fan out to Polish, Slovak, and German. We used AI-assisted translation with human review.

The problem showed up in the first round of user feedback. Slovak and Polish users reported that interpretations "read like a textbook." The translations were accurate but sterile. Tarot interpretation, even the grounded, non-mystical kind, relies on evocative language. Words that prompt reflection, not words that pass a grammar test.

The root cause was our translation pipeline. We were translating from Czech, which meant the target languages inherited Czech sentence structures. German worked fine because the translator restructured aggressively. Polish and Slovak are close enough to Czech that translators preserved too much of the original syntax, producing technically correct but emotionally flat text.

The fix took three weeks. We rebuilt the Polish and Slovak sets by translating from English instead of Czech, which forced more structural distance from the originals. We also added a "read it aloud" test to the review process. If an interpretation sounded like instructions rather than a prompt for reflection, it got rewritten.

Why 20 spreads and not 5

Most tarot apps ship with 3-5 spread layouts. We launched with 20, which felt excessive at the time. The reasoning was a retention hypothesis: if a user tries one spread and likes it, having 19 more to explore gives them a reason to come back.

The data partially confirmed this. Users who tried a second spread type within their first week returned at roughly twice the rate of single-spread users. But the unexpected part was that the spread that drove the most repeat sessions was the simplest one: the single-card daily draw.

Daily card draws now account for the largest share of total sessions. The 20-spread library matters for discovery and first-session engagement, but the daily ritual is what brings people back. We did not plan for this. The daily affirmation feature was added late in the build as a low-effort experiment. It turned out to be the stickiest part of the product.

The daily ritual retention pattern

Here is what the daily draw retention curve looks like compared to spread-based sessions:

| Metric | Daily draw users | Spread-based users |
|---|---|---|
| Day 1 to Day 7 retention | Notably higher | Moderate |
| Avg. sessions per user (first 30 days) | 8-12 | 3-4 |
| Session duration | ~90 seconds | ~4 minutes |
| Most common return time | Morning (7-9 AM local) | Evening (8-11 PM local) |

The daily draw users come back more often, but each session is short. They open the app, draw one card, read the interpretation plus the daily affirmation, and close it. The session is closer to checking a weather app than exploring a content product.

This reshaped how we think about the product. Tarotas started as a spread exploration tool ("here are 20 ways to read tarot"). It became, for its most engaged users, a morning micro-ritual. The content architecture supports both use cases, but the ritual users drive the retention numbers.

Ask-tarot: the feature we almost cut

The ask-tarot mode lets users type a yes/no question and draw a single card. We nearly cut it during the build because it felt too thin. One card, one binary interpretation, done.

We kept it because one team member pointed out that people search "tarot yes or no" in high volume. That search intent alone justified the feature. Once live, ask-tarot sessions showed a behavior we had not predicted: users would ask multiple questions in sequence, drawing 3-5 cards per session. The single-question framing lowered the barrier to engagement, and users treated it as a rapid-fire reflection tool rather than a one-and-done feature.

What the multi-language setup costs us

Running five languages on a single domain (tarotas.com) means every content update is a 5x effort. When we rewrite an interpretation (because user feedback flagged it as unclear or because we improved the reflection framing), that change has to propagate through all five language versions.

We handle this with a localization queue: changes go into the base language (now English for new content), get flagged for translation review, and deploy per language as reviews complete. The lag between an English update and the last language catching up averages about a week. This is not fast, but it is manageable for a product with stable core content that changes infrequently.

The upside of multi-language on one domain is SEO coverage. Each language variant generates its own set of indexable pages, and locale-specific search intents (Czech users searching "tarot vyklad" vs. German users searching "Tarot Deutung") all land on the same product. For a niche product, this approach has outperformed what we expected from organic search.

The build timeline

From first commit to live product with all five languages: about six weeks. The breakdown was roughly two weeks on the card engine and spread logic, one week on the interpretation content for Czech and English, three weeks on the remaining language sets and the translation rebuild described above. The daily affirmation and ask-tarot features were added in week five as side experiments. Both became core parts of the product.

If we built it again, we would start translations from English on day one and skip the Czech-first pipeline entirely. That single decision would have saved us the three-week translation rebuild. We would also invest in the daily draw flow earlier. It turned out to be the product's strongest retention lever, and we had treated it as an afterthought during the build.

---

We build at Inithouse, a product studio shipping AI and consumer products. Tarotas is our tarot reflection app: 78 cards, 20 spreads, 5 languages, no signup required.