-
-
Home: speak or type. Today's review (max 7 words, picked by the Tutor Agent) is ready, so a lesson starts in one tap.
-
A2UI-style card: the Capture Agent picks a word, an expression and a grammar point from a sentence. I tick what to keep.
-
3rd lookup = weak word. Capture hands it to the Tutor, which proposes a swap. Dropping a word waits for my approval (policy.yaml).
-
The wordbook: lookups filed by direction, with examples written by the Tutor Agent. The red sheet hides Spanish to test myself.
-
Two meanings? The Capture Agent asks which one I looked up. My choice becomes the entry, and the Tutor writes an example for it.
-
Agent log: every tool call is checked against policy.yaml before it runs. Allowed, blocked and held calls are all recorded.(for this demo)
-
Prompt-injection demo: two requests a fooled AI might send are blocked by the guard; the normal one is allowed. Nothing runs.
Inspiration
I'm a Japanese exchange student in Spain, learning Spanish through English. I look words up every day and forget them. Translation apps help in the moment, but their history just scrolls away; vocabulary books are too slow to use in a shop or a conversation; chatbots need a prompt every time and never organise anything.
What it does
- Tap and speak (or type). English → Spanish is filed as Couldn't say (a speaking gap), Spanish → English as Didn't catch (a listening gap). The direction is decided by code from the detected language.
- Capture Agent checks the wordbook for the same or a similar entry, then saves it or counts it again. On the 3rd lookup the word becomes a weak word and the Capture Agent hands it over to the Tutor Agent. For sentences it proposes what to learn (a word, an expression, a grammar point) on a card; for a word with several meanings it asks which one you meant.
- Tutor Agent writes an example sentence when it helps and decides today's review (max 7 words): which word to drop for the new one, and why. Dropping a word waits for your approval.
- Lessons (flashcards) and a book-style Wordbook with a draggable red sheet.
- Agent log: every tool call with agent, arguments, allowed / blocked / held, and the reason.
How we built it
- FastAPI + vanilla JS (phone-size UI). Voice streams over a WebSocket to the Gemini Live API; the Cloud Translation API translates at once (no LLM wait).
- Google ADK: a root agent without an LLM routes each job to the Capture Agent or the Tutor Agent (both Gemini). Gemini is called only for decisions: 2 calls per text lookup, 1 per lesson answer, 1 per background review.
- Security: user text is wrapped as data (
<user_utterance>and 7 other data tags).policy.yamllists each agent's tools and limits, and ADKbefore_tool_callbackchecks every call: saved text must equal the input, counts only for search results, no links in examples, max 7 words in today's review, one hand-off per lookup. No agent can delete, change settings or run code. - A2UI-style cards: agents describe cards as JSON from a fixed catalog; the app draws them with textContent, and the server checks every button. Extracted text must be in the input sentence.
- Human approval (
approval_requiredin policy.yaml): dropping a word from today's review on a hand-off waits for the user's approval. - Audit log: every tool call (allowed, blocked or held) goes to
logs/audit.jsonland the in-app Agent log. - Proof: a red-team script sends 16 hijacked-agent tool calls to the real guard: 16/16 blocked. A stage script sends a fooled Gemini's requests to the running app: 2 blocked, the normal one allowed.
Challenges we ran into
Keeping the translation instant while the agents think (translation first, agents after), keeping Gemini calls to a minimum, and making the guard check arguments against facts the server owns, not just tool names.
Accomplishments that we're proud of
A complete working app built in two hours: two cooperating agents, a hand-off, human approval, and 46 passing tests.
What we learned
Least agency works: give each agent a few narrow tools and check every call against facts the server owns, and prompt injection becomes much less dangerous.
What's next for Transtudy
Photo translation, location-based grouping, AI-generated quizzes, and Cloud Run so it runs on a phone.
Built With
- fastapi
- gemini
- google-adk
- google-cloud-translation
- javascript
- python
- sqlite
