Data Scientist, Next Gen Recommendation Systems

Impact.com•New York, NY

2d•Hybrid

About The Position

We're seeking a Data Scientist to help build the next generation of recommendation systems powering our partnership automation platform. Our ecosystem connects a rich set of entities—advertisers, media publishers, creators, products, and consumers—and the relationships between them are where the real value lives. Your work will help surface the right partnerships, the right products, and the right content across this network at scale. You'll contribute to evolving our recommender stack toward a graph-based architecture leveraging semantic embeddings of entities and their relationships, applying cutting-edge techniques in representation learning, graph ML, and retrieval. The system needs to serve recommendations both in batch and real time, respond to dynamic user inputs, drive measurable value for end users across the platform, and remain reliable as the ecosystem grows. This role is hands-on and end-to-end. You'll own modeling and experimentation work for a defined area of the recommendation stack—from problem framing through productionization—in close partnership with Engineering, Product, MLOps, and Business Stakeholders. You're expected to bring (or actively develop) ML engineering chops so you can take a solution from prototype to production, and to be a relentless user of AI coding agents to multiply your output and accelerate iteration.

Requirements

3+ years of experience in data science / applied ML, with a track record of shipping production models that delivered measurable user or business impact.
Strong Python and SQL skills; experience working with large-scale data and distributed compute (Spark/Databricks or equivalent).
Hands-on experience building recommendation or ranking systems—candidate generation, learning-to-rank, retrieval and reranking, or implicit feedback modeling.
Experience with embeddings and representation learning for users, items, content, or other entities.
ML engineering capability (or strong willingness and demonstrated ability to develop it): you can build, ship, and maintain production pipelines—not just prototypes in notebooks.
Strong experimentation skills: designing and analyzing A/B tests, interpreting results, and communicating findings to stakeholders.
Relentless user of AI coding agents in your day-to-day workflow, with a clear sense of where they accelerate you and where they don't.
Insatiable curiosity about new techniques, architectures, and tools—and a track record of teaching yourself things quickly. You read papers, try new tools, and bring ideas back to the team.
Strong problem-solving instincts and the ability to operate with growing autonomy in ambiguous, evolving spaces.

Nice To Haves

Experience with graph-based ML: graph neural networks (KGAT, transformers or similar), graph-aware retrieval, or knowledge graph embeddings.
Experience with modern deep learning recommender architectures (two-tower, sequence/transformer-based recommenders, multi-task ranking).
Familiarity with vector search and vector databases (FAISS, ScaNN, Vespa, Milvus, or similar) and approximate nearest neighbor methods.
Experience with real-time ML serving, feature stores (Feast, Tecton, or equivalent), and low-latency inference patterns.
Exposure to contextual bandits, reinforcement learning, or off-policy evaluation in recommendation settings.
Familiarity with PyTorch/TensorFlow and PyTorch Geometric / DGL for graph workloads.
Familiarity with GCP (Vertex AI, BigQuery, Cloud Run) and/or mature MLOps practices (CI/CD for ML, monitoring, drift detection).
Experience in adtech, martech, e-commerce, or two-sided/multi-sided marketplace recommendations.

Responsibilities

Design, build, and evaluate recommendation models that operate across heterogeneous entities—advertisers, publishers, creators, products, and consumers—and the relationships between them. Frame problems in terms of the partnership graph and apply techniques appropriate to each surface, including candidate generation, ranking, reranking, and personalization.
Contribute to evolving our architecture toward graph-based approaches: learn semantic embeddings of entities and relationships, apply graph neural networks or attention aware graph transformer models where they add value, and build representations that generalize across surfaces and use cases. Stay current with cutting-edge techniques in graph ML, representation learning, and modern recommender architectures, and bring relevant ideas into the platform.
Build models and pipelines that serve recommendations in both batch and real-time contexts. Partner with Engineering on retrieval infrastructure, vector search, feature stores, and low-latency serving patterns. Make pragmatic tradeoffs between model sophistication, latency, cost, and freshness based on the surface and use case.
Own the full lifecycle of your work: data and feature design, model development, evaluation, launch, monitoring, and iteration. Build production-grade pipelines, write code that other engineers can extend, and partner with MLOps on reproducibility, observability, and reliability. Use AI coding agents aggressively to accelerate prototyping, refactoring, debugging, and shipping—we expect this to be a core part of how you work, not an occasional aid.
Design offline evaluation (offline replay, counterfactual evaluation, holdout sets) and online experiments (A/B tests, holdouts, interleaving) to quantify model impact. Apply appropriate statistical methods, recognize common pitfalls in recommender evaluation (position bias, feedback loops, selection effects), and translate results into clear recommendations for product and engineering partners.
Work closely with Product, Engineering, and Business Stakeholders to translate platform goals into measurable model outcomes. Communicate findings, tradeoffs, and recommendations clearly to both technical and non-technical audiences. Document your work so that models, features, and decisions are understandable and reproducible by others.

Benefits

Medical, Dental, and Vision insurance
Office-only catered lunch every Thursday, a healthy snack bar, and great coffee to keep you fueled
Flexible spending accounts and 401(k)
Responsible PTO policy
Mental health and wellness benefit includes up to 12 fully covered therapy/coaching sessions per year, with additional dependent coverage.
Monthly gym reimbursement policy
Restricted Stock Units (RSUs)
Free Coursera subscription
PXA courses
Generous parental leave policy, 26 weeks of fully paid leave for the primary caregiver and 13 weeks fully paid leave for the secondary caregiver.
Technology stipend to help you set up your home office
Monthly allowance to cover your internet expenses.