Job
Description
JOB DESCRIPTION • Strong in Python with libraries such as polars, pandas, numpy, scikit-learn, matplotlib, tensorflow, torch, transformers • Must have: Deep understanding of modern recommendation systems including two-tower , multi-tower , and cross-encoder architectures • Must have: Hands-on experience with deep learning for recommender systems using TensorFlow , Keras , or PyTorch • Must have: Experience generating and using text and image embeddings (e.g., CLIP , ViT , BERT , Sentence Transformers ) for content-based recommendations • Must have: Experience with semantic similarity search and vector retrieval for matching user-item representations • Must have: Proficiency in building embedding-based retrieval models , ANN search , and re-ranking strategies • Must have: Strong understanding of user modeling , item representations , temporal/contextual personalization • Must have: Experience with Vertex AI for training, tuning, deployment, and pipeline orchestration • Must have: Experience designing and deploying machine learning pipelines on Kubernetes (e.g., using Kubeflow Pipelines , Kubeflow on GKE , or custom Kubernetes orchestration ) • Should have experience with Vertex AI Matching Engine or deploying Qdrant , F AISS , ScaNN , on GCP for large-scale retrieval • Should have experience working with Dataproc (Spark/PySpark) for feature extraction, large-scale data prep, and batch scoring • Should have a strong grasp of cold-start problem solving using metadata and multi-modal embeddings • Good to have: Familiarity with Multi-Modal Retrieval Models combining text, image, and tabular features • Good to have: Experience building ranking models (e.g., XGBoost , LightGBM , DLRM ) for candidate re-ranking • Must have: Knowledge of recommender metrics (Recall@K, nDCG, HitRate, MAP) and offline evaluation frameworks • Must have: Experience running A/B tests and interpreting results for model impact • Should be familiar with real-time inference using Vertex AI , Cloud Run , or TF Serving • Should understand feature store concepts , embedding versioning , and serving pipelines • Good to have: Experience with streaming ingestion (Pub/Sub, Dataflow) for updating models or embeddings in near real-time • Good to have: Exposure to LLM-powered ranking or personalization , or hybrid recommender setups • Must follow MLOps practices — version control, CI/CD, monitoring, and infrastructure automation GCP Tools Experience: ML & AI : Vertex AI, Vertex Pipelines, Vertex AI Matching Engine, Kubeflow on GKE, AI Platform Embedding & Retrieval : Matching Engine, FAISS, ScaNN, Qdrant, GKE-hosted vector DBs (Milvus) Storage : BigQuery, Cloud Storage, Firestore Processing : Dataproc (PySpark), Dataflow (batch & stream) Ingestion : Pub/Sub, Cloud Functions, Cloud Run Serving : Vertex AI Online Prediction, TF Serving, Kubernetes-based custom APIs, Cloud Run CI/CD & IaC : GitHub Actions, GitLab CI Show more Show less
