Building Powerful GenAI Apps with Pure Java
summary
Demonstrates a pure‑Java stack that replaces external vector DBs by combining Eclipse Store (object persistence) with GigaMap (off‑heap, lazy‑loaded indexes) and JVector for embeddings, wired through LangChain4j and optionally Ollama for local models; the demo runs a 400 MB embedding model locally, supports bitmap, spatial, Lucene, vector indexes on the same JVM, scales to billions of entries with off‑heap indexing and PQ compression for terabytes, geographic resolution ~1.1 m, uses Panama SIMD, and removes schema drift between object model and vector index.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:38Data as companies' most valuable asset
- 7:24Why vector databases are often overkill
- 10:20Eclipse Store for pure Java persistence
- 12:50GigaMap for high-performance indexing
- 16:28Adding AI with LangChain4j and Ollama
- 18:39What are embeddings and how they work
- 22:41Four indices in one Java collection
- 27:59Wiring the service layer with queries
- 31:51Semantic search unlocks new data insights
- 34:49Geographic search with proximity queries
- 36:34Runtime architecture overview
- 39:02Scaling and performance optimizations
- 39:23Key takeaways and next steps
- 40:12Q&A session
related talks
01h 03mBuilding AI powered Applications with Java
02h 35mBringing GenAI to the Modern Enterprise. A production use-case. In Serverless Java!
50mProduction-Ready GenAI with Open Models for Java Teams
02h 44mUsing Generative AI with Java: Beyond Simple Chatbots
09mMary Grygleski at JNation 2024
52mAI for Java Developers - From Buzzword to Code
53mUnleash the Power of AI Agents with Java Building Intelligent Application
47mBlazingly Fast GenAI App Development Java and Spring AI