🤖
Claims AI Agent & Workflow Automation
Production multi-agent system automating clerical damage claims triage across 1,800+ email threads, cutting processing time by 64% with zero liability risk.
Overview
In logistics operations, transport damage claims involve significant clerical overhead: reading inbound customer emails, identifying container/train tracking numbers, retrieving booking data, and preparing case files.
I built the CargoBeamer Claims AI Agent to automate this triage pipeline end-to-end for the Legal and ECM III teams.
Engineering Architecture
-
Intelligent Field Partitioning:
- Evaluated 55 total case fields:
- 24 auto-filled fields (44%): 15 extracted from inbound emails and 9 enriched from the TMS Booking Data Warehouse.
- 31 human-gated fields: Reserved strictly for human legal and claims specialists (0€ autonomous liability, zero autonomous payout rights).
- Evaluated 55 total case fields:
-
Noise Triage & Thread Bundling:
- System bundles ~4 correspondence emails into a single case file, eliminating duplicate entries across 1,800+ inbound threads.
- Batch executions run 2x daily in ~4 minutes with +38% surge headroom.
-
Motion Graphics Explainer System:
- Engineered a 108-second broadcast-quality 1080p motion graphics explainer video using React and Remotion (
claims-agent-video), featuring automated narration and dynamic burned-in kinetic typography.
- Engineered a 108-second broadcast-quality 1080p motion graphics explainer video using React and Remotion (
Impact
- Slashed clerical intake time by 64% (from 25 minutes down to ~9 minutes per claim).
- Recovers ~16 hours per month (~192 hours/year) of high-value legal specialist time.