Harvey Taxi Service LLC

The live prototype uses Meta Muse Spark 1.3 through the Meta Model API to review fictional accessible transportation booking documents, select tools, record exact source quotations, and prepare a cited review memo. Independent application checks validate evidence, fare calculations, and required booking information. Missing information stops the workflow for clarification; a completed review awaits human acknowledgment. Saved run records and API receipts support inspection of the model’s actions. We also completed a separate Muse Glimmer 30B LoRA fine-tuning experiment on Fireworks AI using examples of our demonstration fare and escalation rules. Fine-tuning is experimental; deterministic application checks enforce fare correctness. The review console is hosted on Modal. Curb-clearance values and voice transcripts are supplied demonstration inputs. Live SAM image measurement and Meta audio transcription are not integrated. The prototype does not dispatch vehicles, process payments, or provide clinical clearance. Project overview: https://williebee615.github.io/harvey-taxi-voice-agent/docs/ Source and setup: https://github.com/Williebee615/harvey-taxi-voice-agent