Case study
Freightline — RAG Copilot for Logistics Ops
A retrieval assistant that answers dispatcher questions from 40k pages of carrier contracts, tariffs, and SOPs — with citations.
01The problem
Freightline's dispatch team fielded hundreds of daily questions about carrier agreements, accessorial charges, and routing rules. Answers lived in 40,000 pages of PDFs across three drives. Senior staff spent up to two hours a day being human search engines, and mistakes on tariff terms were costing real money on disputed invoices.
02What we built
We built a retrieval pipeline with structure-aware chunking that respects contract clause boundaries, layered search with a reranking stage, and an answer layer that refuses to respond without a source. Every answer cites clause and page. An evaluation harness with 400 curated Q&A pairs gates every release; retrieval quality is tracked per document family on a live dashboard.
03The outcome
First-response time on contract questions dropped from 14 minutes to under 30 seconds. The assistant answers 78% of questions without escalation, and every escalation lands with the right context attached. Invoice-dispute losses tied to misquoted terms fell 40% in the first quarter.
