The problem
I enjoy learning languages, but I was dissatisfied with the apps available to me. Most either prescribed the material, reduced practice to a short daily game, or covered one skill without giving me a place to connect it to the others. That did not match how I wanted to study, especially around the B1 and B2 levels where reading and independent practice become much more important.
I built Blueprint for my own workflow first. It is meant for an intentional study session of thirty minutes or an hour, not something to tap through for five minutes on the train. The learner brings in a text or video, works through it carefully, saves difficult vocabulary, generates exercises, speaks with a tutor, writes, and returns for review.
What I built
Blueprint brings the four parts of language learning into one workflow: reading, writing, listening, and speaking. A learner can import a URL, PDF, passage of text, or YouTube video and turn it into material they can actively study. Words in the reader can be looked up and saved, the content can become grammar and writing exercises, and Gemini Live provides a place to practise speaking.
The vocabulary system is the part I care about most. Each word keeps a mastery score based on review history, with exercises for both recognition and active recall. That makes it possible to see which words are becoming familiar and which ones keep causing trouble, rather than treating every saved word as equally learned. AI supports the process by explaining, generating practice, checking written work, and holding conversations. It complements the learner's effort instead of doing the learning for them.
Everything you want to learn, in one place
The library is where a study session starts. Import a source and Blueprint prepares the content in the background so it is ready to read.
- Import from a URL, PDF, raw text, or YouTube link
- Lessons are grouped by language and tagged for curriculum building
- Choose material that you genuinely want to understand
Decisions and tradeoffs
Token-based reader, not plain text
Tokenizing every imported document up front adds some processing time, but it makes each word interactive afterward. Lookups, mastery tracking, and flashcard context all use the same token layer.
SM-2 instead of an LLM scheduler
Review scheduling uses the SM-2 algorithm rather than asking an LLM what to review. It is deterministic and predictable, leaving the AI for tasks where generated language is useful: explanations, conversation, exercises, and feedback.
Cost observability from day one
Every Gemini call is metered by user and feature. Knowing what each part of the product costs is necessary if Blueprint is going to grow beyond a tool I use by myself.
Designed for deliberate study
Blueprint does not try to turn language learning into a string of tiny games. It asks the learner to choose material, spend time with it, produce their own language, and come back for the reviews that need attention.
What I learned
Building the features was only part of the work. Using Blueprint during my own study showed me that small interruptions matter: an extra step before reading or a review screen with too much information can break a focused session.
It also made the right role for AI clearer. Generated language is useful for explanations, feedback, and conversation, but progress tracking and review scheduling are easier to trust when they remain deterministic.
Result
Blueprint is nearly feature-complete and is now in testing. The full study loop works across reading, vocabulary, flashcards, writing, grammar, and voice conversation, with support for CJK features such as pinyin and furigana. I use it as part of my own language study as often as I can, which is helping me find the small problems that only appear during real, repeated use. The next step is to keep refining that daily experience before opening it up more widely.
