Staff Backend Engineer, Recommender Systems

As a Staff Backend Engineer at Raya, you will be the technical architect and hands-on builder for our recommendation ecosystem. You’ll build and evolve sophisticated, multi-stage retrieval and ranking systems, bridging applied ML/AI with production backend engineering to deliver algorithms that are both performant and intelligent. You will join at a pivotal moment as we scale our recommendation systems to support growth and increasingly complex marketplace dynamics. Responsibilities

Architectural Leadership: Own the end-to-end architecture of Raya’s recommendation services while remaining deeply hands-on in implementation. 

Hands-on Implementation: Design and ship systems that handle cold-start problems, real-time user signals, exposure balancing, and large-scale feature lookups.

System Evolution: Evolve our ranking systems toward scalable multi-stage architectures, including embedding-based retrieval and graph-aware ranking where appropriate.

Cross-Functional Influence: Act as the primary technical liaison between Data Science, Product, and Infrastructure. Translate complex algorithmic requirements into scalable backend services.

Mentorship & Excellence: Elevate the engineering bar across the organization. Conduct deep-dive design reviews, establishing best practice standards for backend patterns, and mentor Senior Engineers in recommender systems best practices.

Operational Stewardship: Ensure the reliability of mission-critical recommendation loops. Optimize for low-latency inference and high-availability, even during peak global traffic. 

Ambiguity & Tradeoffs: Operate in evolving problem spaces where objectives must balance short-term engagement, long-term retention, and marketplace health.

Experimentation: Partner with Product/Data Science to implement offline + online experiments.

Qualifications

Education: Bachelor’s degree in Computer Science, Engineering, Mathematics, or equivalent real-world expertise building and operating production recommendation or ranking systems.

Experience: 8+ years of software development experience, with at least 3 years focused specifically on Recommender Systems in a production environment.

RecSys Mastery: Deep practical experience with recommender approaches like collaborative filtering, content-based filtering, and hybrid models. Experience with two-stage architectures (Candidate Generation & Ranking). 

Infrastructure Skills: Expert-level proficiency in Golang, Node.js, or Python. Experience building or operating high-throughput discovery, search, or recommendation systems in production.

Data Fluency: Advanced knowledge of Postgres, MongoDB, and ElasticSearch/OpenSearch, specifically regarding performance tuning for high-concurrency discovery features.

System Design: A history of shipping platforms that have scaled to millions of users. You should be comfortable discussing the trade-offs between consistency, availability, and latency.

A/B Testing: Experience designing and implementing A/B tests in marketplace or interference-prone environments.

What Sets You Apart

Marketplace Intuition: You understand that ranking people is fundamentally different from ranking content. You’ve worked in environments (dating, social, marketplaces, ride-sharing) where exposure affects behavior, and you design with fairness, liquidity, and user perception in mind.

The "Product Engineer" Mindset:  You bring strong product judgment to technical decisions, protecting serendipity, privacy, and user trust while shipping measurable improvements.

Systems Builder: You build durable internal abstractions, tooling, and documentation that make future iteration faster and safer.

Algorithmic Intuition: You understand the math behind ranking models and can identify bias, feedback loops, and unintended system behaviors before they become production issues.

Strategic Pragmatism: You optimize for shipping measurable impact over technical novelty. You know when to apply a simple heuristic and when to deploy a complex model.

Bias Toward Shipping: You build quickly, learn from production signals, and iterate with discipline rather than over-optimizing prematurely.