11 hours ago
Base Salary
$230k - $322k/yr
Responsibilities
- Define the technical direction and multi-year roadmap for ads retrieval modeling.
- Design, develop, and launch candidate-generation and retrieval models for campaigns and ads across Reddit’s advertising surfaces.
- Apply two-tower architectures, representation learning, embeddings, sequence models, graph-based methods, and deep learning techniques to retrieval problems.
- Improve objectives, labels, sampling strategies, hard-negative mining, feature design, embedding generation, candidate filtering, and retrieval depth.
- Work with approximate nearest-neighbor and vector retrieval systems while balancing recall, relevance, freshness, diversity, coverage, latency, and cost.
- Establish evaluation practices connecting retrieval metrics to downstream advertising and user outcomes.
- Lead offline analyses and online experiments and translate results into modeling iterations.
- Partner with ranking, ads platform, auction, measurement, and product teams to integrate retrieval models into the ads funnel.
- Write design documents, review code and model changes, and improve modeling, testing, observability, and production ownership.
- Mentor ML engineers and grow team expertise in retrieval, recommendation, and representation learning.
Requirements
- 7+ years of industry experience, including substantial experience building and shipping applied ML products.
- Deep experience with information retrieval, candidate generation, recommender systems, ranking, or related relevance problems.
- Strong understanding of DNNs, embeddings, two-tower or dual-encoder models, approximate nearest-neighbor search, and multi-stage retrieval.
- Deep experience training, evaluating, debugging, and deploying deep learning models using TensorFlow, PyTorch, or similar frameworks.
- Demonstrated ownership of ML projects from problem framing and data preparation through evaluation, experimentation, production launch, and iteration.
- Strong command of experimental design and model evaluation, including relationships between offline retrieval metrics and downstream business and user metrics.
- Experience with large-scale behavioral, contextual, or content datasets and complex feature pipelines.
- Strong software engineering fundamentals and ability to write clear, reliable, maintainable production code.
- Technical leadership experience setting direction, leading complex projects, influencing partner teams, and mentoring engineers.
- Excellent written and verbal communication skills.
- Preferred: experience with ads retrieval, ad serving, recommendation, search relevance, marketplace optimization, sequential or graph modeling, multimodal signals, ads marketplaces, or applied ML and ranking publications, patents, or industry contributions.
- Preferred: experience connecting retrieval improvements to ranking, auction, conversion, revenue, or user-experience outcomes and experience with RNNs or Transformers.
Benefits
- 100% remote opportunity, with hybrid or onsite work available at offices in New York, San Francisco, Los Angeles, and Chicago.
- Comprehensive healthcare benefits and income replacement programs.
- 401(k) with employer match.
- Global benefits supporting workspace, professional development, caregiving, and lifestyle needs.
- Family planning support, gender-affirming care, and mental health and coaching benefits.
- Flexible vacation and paid volunteer time off.
- Generous paid parental leave.
- Base salary is eligible for additional equity and, depending on the position, commission; compensation figures are excluded from this benefits summary.
Tech Stack
PyTorchTensorFlow
Categories
About Reddit
Reddit builds a social news and discussion platform organized into user-created forums (“subreddits”) for consumers, creators, and communities. It monetizes primarily through advertising tools for brands and self-serve advertisers, plus a Premium subscription; developers can access data via its API. Founded in 2005 and headquartered in San Francisco, Reddit became a public company on the NYSE in 2024 and hosts over 100,000 active communities.