LLM Supplier Entity-Matching Platform on GCP
Client: Supplier-data integration partner
An LLM entity-matching platform on GCP that matches supplier records in real time, accepting a match only when Gemini, Claude and GPT agree.
The Challenge
Supplier and vendor records from many systems had to be matched to a third-party supplier-data platform in real time. Name-based rules gave too many false matches, and runs that took several hours failed with no way to resume.
Our Solution
Built about 90% of the platform. It started as an AWS build (Amazon MQ, Lambda, MWAA, ECS Fargate, Bedrock, Redshift) and we led its 2026 rebuild on GCP with RabbitMQ, Cloud Composer, GKE and Snowflake. Matching uses three LLMs (Gemini, Claude, GPT) and only accepts a match when they agree.
Outcomes
Running in production on GCP
Three-model consensus matching reduces false positives from any one model
Incremental checkpointing lets multi-hour runs resume after a failure
Idempotent event handling, 5 containerised stages, CI/CD across 5 services
Technologies
Have a Similar Project?
Tell us about your data or AI challenge and we'll show you how we would approach it.