Team unjargon agent is an event-driven, privacy-first Collaborative Partner that turns jargon encountered with Claude Code and Codex into reusable team knowledge.
A local bridge sends only conservative candidate terms and a source category—never transcripts, prompts, paths, or session IDs. For every event, the agent deduplicates terms, retrieves voluntarily shared team definitions, automatically aligns known terms, and queues genuinely new ones for a learner. Its on-screen Autonomous decision log makes that work auditable while persisting only the term and routing decision.
A learner can request a concise explanation with Google ADK and Gemini 3.5 Flash-Lite, then decide whether to share it. That explicit feedback changes the next autonomous run: the same term is subsequently aligned from team memory without another model call. Teams can also export and import their shared glossary as Markdown; only term headings and the first definition line are retained.
The deployed application uses Cloud Run, Vertex AI Gemini 3.5 Flash-Lite, Google ADK, and Firestore. The public repository includes reproducible setup, privacy boundaries, tests, architecture diagram, four-minute demo script, and pre-existing-work disclosure.
Built With
- cloud-run
- fastapi
- firestore
- gemini-3.5-flash-lite
- google-adk
- vertex-ai
Log in or sign up for Devpost to join the conversation.