Hacktoberfest Hack Day Linz x Young AI Leaders Linz
Join us at JKU Linz on October 30 for Austria's exclusive MLH Hacktoberfest Hack Day, a free, beginner-friendly mini-hackathon where students and professionals build open-source AI prototypes without the sleepless nights. Team up to experiment with partner tools from Google Gemma, Lovable, and ElevenLabs, receive hands-on mentorship, and compete for unique Major League Hacking swag and prizes.
Tripwire
Practice spotting phishing emails and social-engineering calls with Gemma-powered scenarios, live voices, and evidence-based feedback.
About the project
Tripwire is a security-awareness training simulator for employees of the fictional Donauwerk Logistik GmbH in Linz. In a five-scenario round, trainees inspect realistic emails and talk to callers, deciding whether to comply with a legitimate request or report a scam. The debrief explains the decision with red flags, concrete evidence, and safe signals, and the round ends with a score.
The Lovable-built interface provides a simulated inbox, link-destination inspection, in-app landing pages, attachment previews, and a live call screen. Scenarios use reserved .example domains, and links and attachment interactions are simulated rather than directing trainees to real phishing sites.
Gemma 4 (gemma-4-26b-a4b-it) is used in two distinct ways: a Python generator creates structured, labelled email and call scenarios with validation and correction retries; and a Rust relay streams scenario-specific conversational replies to ElevenLabs during live calls. Generated scenarios are committed as fixtures, not generated afresh for each round. The repository currently contains 20 generated scenarios alongside three hand-written fixtures.
The voice integration uses ElevenLabs Agents with a custom-LLM connection to the Rust backend. The backend supplies the caller persona and goal from its scenario data, forwards conversation and tool requests to Gemma, and streams the response back. Speech input and output are handled by ElevenLabs. The call interface lets the trainee talk, hang up, and explicitly choose a decision; a disconnection is not treated as proof of compliance.
The backend uses Rust, Axum, and Tokio, with SQLite event logging and an optional io_uring HTTP server. Gemma can run locally through llama.cpp/llama-server; the documented deployment uses self-hosted Gemma 4 26B A4B Q4_0 on CPU. The backend and generator also support Gemini API hosting of Gemma via an OpenAI-compatible API. This does not imply use of an OpenAI model. The frontend uses React, TypeScript, TanStack Start, and Tailwind CSS.
The source repository is MIT licensed. This draft is based on the current main README and source, with the user-confirmed application name Tripwire. The Lovable preview includes newer voice-routing and microphone-meter changes that are not yet reflected in the repository frontend snapshot. No demo video is attached yet.
Built with
AI tools
Prize Categories Entered
- Submitted
Best Use of Gemma 4
- Submitted

Best Open-Source AI Project
- Event prizeSubmitted
Best Use of Lovable
- Event prizeSubmitted
Best Use of ElevenLabs