18,000 líneas de Python que leen correos, calculan rutas y generan cotizaciones sin intervención humana. LoRA fine-tuning propio al 94.1% quality. 142 quotes/día en producción. 18,000 lines of Python reading emails, computing routes and generating quotes without human intervention. Custom LoRA fine-tuning at 94.1% quality. 142 quotes/day in production.
Cada día un dispatcher típico recibe cientos de RFQs por email — direcciones, equipo, fechas, todo en lenguaje natural caótico. Tiene que leerlos, extraer datos, validar en el TMS, calcular ruta y rate, y responder. Manualmente. Un por uno. Y los buenos shippers no esperan. Each day a typical dispatcher receives hundreds of RFQs by email — addresses, equipment, dates, all in chaotic natural language. They must read each one, extract data, validate in the TMS, compute route and rate, and respond. Manually. One by one. And good shippers don't wait.
Kamila lee el correo entrante, identifica que es un RFQ, extrae lanes/equipment/fechas, valida en el TMS, calcula la ruta óptima y genera el quote — todo sin tocar nada. El dispatcher solo aprueba antes de enviar. Kamila reads the incoming email, identifies it as an RFQ, extracts lanes/equipment/dates, validates in the TMS, computes the optimal route and generates the quote — all hands-off. The dispatcher only approves before sending.
Microsoft Graph + IMAP polling. Identifica RFQs entre ruido, threading inteligente, deduplica, prioriza por shipper tier.Microsoft Graph + IMAP polling. Identifies RFQs in noise, intelligent threading, dedup, prioritizes by shipper tier.
LoRA fine-tuning propio sobre 4K+ RFQs reales. Extrae lanes, equipment, accessorials, fechas, special handling. 94.1% quality on holdout.Custom LoRA fine-tuning on 4K+ real RFQs. Extracts lanes, equipment, accessorials, dates, special handling. 94.1% quality on holdout.
18 endpoints TILT/Bridgeway. Validación de lanes, rate logic propia, multi-stop detection, dead letter triage. Human-in-loop console para aprobación.18 TILT/Bridgeway endpoints. Lane validation, custom rate logic, multi-stop detection, dead letter triage. Human-in-loop approval console.
Ejemplo real (anonimizado) de un RFQ procesado por Kamila el mes pasado. Tiempo desde email recibido hasta quote listo: 28 segundos. Real example (anonymized) of an RFQ processed by Kamila last month. Time from email received to quote ready: 28 seconds.
SaaS multi-tenant gestionado, licencia on-prem para tu propia infraestructura, o build-out completo custom. Todos incluyen fine-tuning sobre tus propios RFQs. Managed multi-tenant SaaS, on-prem license for your own infrastructure, or full custom build-out. All include fine-tuning on your own RFQs.
Volumen de RFQs alto, lanes complejas, o un TMS difícil — empezamos con una llamada de 30 minutos y un quote en 5 días.High RFQ volume, complex lanes, or a difficult TMS — we start with a 30-minute call and a quote in 5 days.