Captur: Observability-First Mobile ML Inference for Better Customer Confidence

Captur builds a mobile SDK that brings real-time image recognition and actionable feedback directly into customers’ apps, running complex machine learning models entirely on device without cloud inference. This architecture delivers privacy and performance, but also creates unique challenges when it comes to observability and debugging, especially as crashes can originate from the host app rather than the SDK itself.

In this session, Sumanas Sarma and Justin Powell, Captur’s CTO and Head of Mobile, walk through how the SDK is built and why traditional reactive crash debugging, which involved days of investigation, was unsustainable at scale. They’ll explain how Captur moved to a proactive approach using Datadog RUM to isolate SDK issues in real time, reduce irrelevant session noise by controlling sampling and filtering, and reach out to customers before they report problems.

The talk will also cover how Captur leverages Datadog MCP Server and AI-driven code assistance to drive work based on real usage data, improving both observability and customer trust.

Attendees will come away with practical patterns for using real-user monitoring and observability data to isolate hard-to-find failure modes, reduce time spent on reactive debugging, and improve both product quality and customer communication in any environment where visibility into user-facing behavior matters.