Self-Hosted LLM Fine-Tune & RAG

LoRA, QLoRA, RAG, Qdrant, Chroma, pgvector, BGE, Qwen3 embeddings. The decision tree, the dataset shape, the agency brief, and the red flags that kill bad proposals.

• The 8-question fine-tune-vs-RAG decision tree, the 2026 LoRA hyperparameter starting point, and the hardware cost reality in one chart
• The 5-library decision matrix, the vector database decision matrix, and the embedding model leaderboard families with per-token cost
• The hybrid retrieval pipeline, the agency brief verbatim, the 9 red flags, and the RAGAS evaluation metrics for any production RAG