This is a submission for the Gemma 4 Challenge: Build with Gemma 4
What I Built
LyricLens is an AI-powered lyrical forensics tool that helps listeners uncover the poetry, metaphors, and storytelling often missed in fast-paced music consumption.
Music listeners often connect emotionally with songs without fully understanding their deeper meaning. LyricLens automates lyric analysis, delivering instant insights into metaphors, themes, and storytelling.
LyricLens creates an immersive, atmospheric experience where users can get a detailed analysis covering core themes, emotional arcs, and literary devices.
Demo
Code
https://github.com/msoftware/lyriclens
How I Used Gemma 4
The core "intelligence" of LyricLens is powered by the Gemma 4 26B (A4B-IT) model. This specific variant offers a balance between reasoning capabilities and inference speed. The Gemma 4 26B architecture allows it to catch subtle metaphors and "read between the lines" providing a high-quality human-like analysis.
For the radar chart to work, the model needs to output valid JSON consistently alongside its textual analysis, is reliable at following complex system instructions and generating structured data.
I needed a model that could not only translate but reason in the target language to ensure the tone and nuance of the analysis remained consistent.
Using the A4B model ensures that even a model of this size stays responsive and efficient for real-time analysis, significantly reducing latency and compute costs compared to dense alternatives.
