All activity
PatientGlue helps medical practices cut no-shows & save admin time with secure, two-way SMS integrated into DrChrono. Ai integrated, quick setup, risk-free pilot, happier patients & smoother operations.
Patient GlueHappy patients stick around
This repo is for the demonstration of TSCE principles. - AutomationOptimization/tsce_demo
TSCE is an open source framework that increases the accuracy and reproducibility of LLM's and AI agents.
Run more than 4000 test prompts, noted an uplift of +10 - +30pp
Two-Step Contextual EnrichmentTSCE is model-agnostic and increases LLM accuracy +20-30pp
kaleb cadenheadleft a comment
Been building AI agents and agentic workflows since early 2023. Had a repeat issue where I'd find I could incorporate AI into a workflow in novel ways, but the output was too unreliable to use it at scale. Went back to the drawing board, read up on some other methods similar to this, and thus TSCE was born!
Two-Step Contextual EnrichmentTSCE is model-agnostic and increases LLM accuracy +20-30pp
kaleb cadenheadleft a comment
https://github.com/AutomationOptimization/tsce_demo Two-Step Contextual Enrichment (TSCE): first make an LLM spill a high-temperature, symbol-heavy anchor and then force another for the final answer to decode that anchor at T ≈ 0.1. On 300 GPT-3.5 prompts it jumped from 49 % → 79 % accuracy (+30 pp) and cut rule-breaks to 0/300, all for ~1.2 × token cost
🔥 Roast my idea: drop your ideas and get brutally honest feedback 🔥
Aaron O'LearyJoin the discussion