Repository navigation
NeonLib — an Android knowledge library powered by EmbeddingGemma #824
gorrc
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi Gemma community!
I'd like to share NeonLib, an Android app I've been building that uses EmbeddingGemma 300M through LiteRT for on-device document retrieval.
Google Play:
https://play.google.com/store/apps/details?id=com.neonbreak.neonlib&hl=en
From documents to searchable knowledge
NeonLib brings notes, imported documents and text recognised from photos or scanned PDFs into a personal library.
Users can ask a question, retrieve relevant passages and open the original source. Optional answers are generated by a separate downloaded local language model or a selected online AI provider.
EmbeddingGemma handles retrieval, not answer generation. Local retrieval does not require a cloud embedding API.
Under the hood
NeonLib also supports sharing private collections with optional prepared vectors. The receiving app checks model/tokenizer fingerprints, preprocessing settings and exact chunk text before reusing those vectors. Incompatible vectors fall back to local indexing.
Looking for feedback
I'm particularly interested in Croatian and English retrieval, mobile performance, and consistency between desktop-prepared document vectors and Android-generated query vectors.
Initial functional checks work, but broader retrieval benchmarks and device performance measurements are still work in progress.
I'd love to hear from others building with EmbeddingGemma:
Thanks to the Gemma and Google AI Edge teams for making this possible!
All reactions