EverydayGPT: Confidence-Gated Routing for Efficient and Safe Hybrid GPT-RAG Conversational QA

In three linesEverydayGPT introduces Confidence-Gated Routing (CGR) for conversational RAG systems. A 205M-parameter GPT trained on 10B tokens avoids costly GPT pathway (~5.9s) for 85% of queries by resolving them via fast RAG extraction (~45ms), reducing latency 120x. F1=0.226 on 500-question benchmark, with 6.3x mean latency reduction.

Summary generated by Claude — human-verified