Field notes

The Aspexilary
Technical Blog

LLM fine-tuning, RAG architecture, and on-premises AI deployment for regulated industries.

fine-tuning llm java gemma
Gemma 4 31B Beats Our Qwen Models on Enterprise Java
How we fine-tuned Google's Gemma 4 31B on enterprise Java codebases and achieved a 61% improvement over our previous best model.
April 4, 2026
fine-tuning llm java
Fine-Tuning a Code LLM on 73K Java Enterprise Examples
How we trained a Qwen 2.5 Coder 14B model on WildFly, Spring, Kafka, and Hibernate codebases — LoRA, Q4_K_M quantization, and lessons learned.
March 1, 2025
post fine-tuning java llm
Fine-Tuning a Java LLM: From Dataset to Deployment
How we fine-tuned Qwen 2.5 Coder 14B on 73,910 Java enterprise examples using LoRA, quantized to Q4_K_M, and achieved 2x performance over DeepSeek on stream processing tasks.
February 15, 2025
post rag compliance infrastructure
RAG for Regulated Industries: IBC, OSHA, and Retrieval That Actually Works
Building a production RAG pipeline over regulatory corpora using BGE-Large embeddings and Qdrant, with the design decisions that make compliance retrieval different from general-purpose RAG.
January 28, 2025