Fire TV Catalog powers the content discovery experience for tens of millions of customers worldwide. At its core, the catalog must answer a deceptively hard question: is this the same movie, show, game, or clip across dozens of providers? The Matching team owns entity resolution — the system that groups provider items referring to the same real-world content into unified clusters. This is what makes it possible for a customer to see a single tile for a movie with rent, buy, and subscription options from multiple services side by side, rather than a fragmented wall of duplicates. Our matching pipeline processes tens-of-millions of records incrementally across single-host and distributed architectures, maintaining low end-to-end latency while handling complex deduplication across hundreds of content providers. We already leverage LLMs in production for match inference — using large language models to reason about whether two items represent the same content in cases where traditional signals are ambiguous or insufficient. As Fire TV expands into short-form video, live sports, and new entity types, the matching problem is evolving — traditional title/year/ID signals don't always apply, temporal and event-based matching introduces new dimensions, and the scale and diversity of content continues to grow.

Designing and evolving matching algorithms that combine ML scoring, LLM-based inference, deterministic rules, external ID linking, and guard rails to produce high-precision clusters at catalog scale

Expanding and improving our production use of LLMs for match decisions — optimizing prompt strategies, evaluating model performance, managing cost/latency trade-offs, and identifying new areas where LLM reasoning can replace or augment heuristic logic

Driving the expansion of matching capabilities to new content types (short-form, sports events, live content) where existing signals and heuristics break down

Owning the end-to-end quality of match decisions — precision, recall, and the customer-visible impact of getting it wrong (merged content that shouldn't be, or duplicates that persist)

Leading technical design, code reviews, and operational excellence for a team of engineers

Championing AI-assisted development practices across the team — setting the standard for how engineers use AI tools to write, review, test, and debug code more effectively

Partnering with ingestion, curation, and publication teams to ensure matching integrates cleanly into the broader catalog pipeline

Defining metrics, monitoring, and debugging tools that make match quality observable and actionable