How a global telecom scaled agentic AI with synthetic data

An Asian telecom leader serving 30+ million subscribers was building a personal-assistant app, but early agentic AI prototypes suffered from poor context retention, generic responses, unreliable tool use, and no scalable evaluation metrics beyond manual "vibe checks." Snorkel built modular data-curation pipelines spanning the development lifecycle, including programmatic generation of multi-turn conversations, over 20 task-specific quality evaluators, and a custom framework with 60,000+ synthetic training/evaluation datapoints, then fine-tuned open-source models to replace proprietary APIs. The result was an 8% performance improvement over baseline Llama models and a reported +35 point gain in function-calling accuracy, achieved in about two months while also cutting API costs.

The customer is anonymised in Snorkel AI’s own case study — not withheld by us.

Snorkel AI is a data-development and evaluation company whose programmatic labeling tools build training and benchmark datasets used to stress-test whether a model approach is viable before a full build.

This sits in the Design & Stress-Test stage of the AI use-case lifecycle: Named examples from the tools pharma teams use to scope, prototype and benchmark an AI use case before committing to build it — data-catalog, sandbox and evaluation platforms.