{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/5f980ac26b8c43de90718274529dfebd\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/5f980ac26b8c43de90718274529dfebd-1519d511f89c6b1a.gif","duration":304.29,"title":"LLM Sentiment Evaluation Harness Overview","description":"This Loom presents an LLM Sentiment Classifier Evaluation Harness and its engineering architecture. It loads customer transcripts, runs LLM evaluations concurrently across multiple providers, normalizes predictions, exports CSV and JSON reports, and programmatically flags ground truth data errors from specific anomalous IDs including 15, 59, 19, and 47. Key design elements include a modular structure following SOLID and DRY principles, robust parallel processing with dynamic rate limiting cooldown, graceful termination, and device independence. The speaker also outlines core components such as a config file, evaluator using a thread pool executor, loaders, metrics for class precision, recall and F1 score, reporters for CSV and JSON output, and a main CLI entry point, with an additional intention to build a front end using FastAPI or similar frameworks for easier report generation from uploaded CSV files."}