Case Study: Montreal-Based AI Company Achieves Faster, 94.77% Accurate Text Annotation with Wrk

Complex Text Annotation for a Leading AI Company

Montreal-Based AI Company, headquartered in Canada’s AI hub, needed a way to clean up messy incoming data and accurately annotate complex text in multi-level, nested formats for model training. The company relied on Wrk to help automate what had largely been a manual process, using Wrkflow and Wrk Actions to standardize and structure diverse input files before classification.

Wrk implemented a 5-step Wrk Actions Wrkflow that extracted and cleaned data, ran an initial NLP pass, and then classified levels 1, 2, and 3 with filtering steps in between. The result was structured output at 94.77% accuracy on about 4 million text items, with all-in cost of $0.87 per 3-level classification and 2X classification throughput, delivering roughly 200% faster turnaround than the client’s in-house process.