__________________________________________
Adaptive Inference Suppression · Patent pending · USPTO #64/006,312
✅ Verified: 830 queries · 71.8% API savings · 12 GPU calls out of 830
📹 Recorded installation & benchmark run
Full install via
proof_install.sh → benchmark → live routing visible in terminal. No edits.📊 Benchmark results — 830 queries, Groq API backend
830
Total queries
71.8%
API calls saved
12
GPU / LLM calls
596
Served from cache & KB
🔍 Routing layer breakdown
How it works: PRL3 sits between your application and your LLM provider. Incoming queries are routed through a multi-layer stack — exact cache, intent classifier, PHP generator, and a self-learning PRL knowledge base — before any call reaches the LLM. Only queries that cannot be resolved locally are forwarded to the API. The system learns continuously: each verified LLM response is absorbed into the knowledge base, improving future hit rates automatically.