* Media (video & image) platform: end-to-end media ingestion infra (uploading, transcoding, and delivery/serving); visual signal computation infra; media optimization.

* Content Acquisition/Ingestion

Link scraping/crawling and parsing; link clustering, propagation & aggregation between pin links and images; high quality localness and trustworthy signals; crawled rich pin creation/ingestion infra; localized canonical pin ranking etc.

ML Signal/Feature Platform team owns critical data infrastructure & central platform to boost and scale signal/feature development efforts across entire company.

I'm the founder, architect and TL of a novel "micro-data platform" (internal codename "Galaxy") which provides:

• annotation-based generic DataFlow Java API (DSL) with extension for incremental processing

• managed signal registry

• state management and execution orchestration

• ownership model and automated monitoring/alerting

• fully-automated metadata (lineage, data coverage & distribution, delay etc) generation, collection and serving

• strong governance, signal discovery, and signal life-stage dev flow

• rich tooling (command line & UI)

In addition, I server as uber-TL who's responsible for an umbrella of projects:

* User Signal Platform V1/V2: company-wide central platform for realtime user interest & intent understanding

* Discovery-indexing data pipelines and its migration to "Galaxy"

* Engagement event-driven (Kafka consumer based) common infrastructure for realtime signals

Built-from-scratch a unified user signal platform of development framework, infrastructure, offline data access and online serving. I innovated a novel data processing+serving lambda architecture, spreading incremental computation across batch, real-time and serving-time. User signal platform provides user features/signals to power Homefeed Ranking, Ads Ranking and other candidate retrieval/generation systems at Pinterest.

Alongside the project, I introduced modern Dependency Injection technology to Pinterest and laid out a modularized framework for building processing/serving applications. I built a highly efficient real-time event processing framework/infra that processes xxx K/sec event qps with 25x higher efficiency than old infra.

Proposed and drove framework/platform initiative for user modeling/feature engineering in Homefeed recommendation system. Built a unified user signal framework and service, and fully integrated with entire ML infra of Homefeed including model DSL, offline training/eval pipeline, and online scoring/serving path.