Wizard - Machine Learning Engineer – Search & Retrieval Systems

• Own and evolve the hybrid search pipeline – lexical retrieval, dense vector search, reciprocal rank fusion, and multi-stage reranking on Elasticsearch • Build and train adaptive retrieval models – lightweight classifiers and ranking models that configure search behavior per query, per category, per context (source routing, per-attribute boost prediction, filter mode decisions) • Design and productionize the learning-to-rank system – from feature engineering through model training (LightGBM, ONNX) to production deployment and A/B evaluation • Build the search feedback loop – instrument and integrate behavioral signals (CTR, conversions, add-to-cart) into ranking and retrieval as features for LTR, reward signals for adaptive retrieval, and inputs for search-side personalization • Build the business and ordering layer – separating organic relevance from sponsored/partner placement with quality gates, slot allocation, campaign configuration, and an auction-style approach as the system matures • Own the offline enrichment pipeline – LLM-based product enrichment at scale, data quality monitoring, and index management • Instrument and evaluate everything – bulk evaluation pipelines, per-category metric tracking, regression detection, experiment analysis • Integrate query understanding outputs into retrieval – translating extracted attributes, intents, and constraints into filters, boosts, and retrieval strategy decisions • What Success Looks Like • You ship ranking and retrieval improvements that measurably move product quality metrics – accuracy, NDCG, latency • The search pipeline adapts its behavior based on query context rather than relying on static configuration • You own systems end-to-end: from the training data pipeline through model training to production serving and evaluation • You build infrastructure that other engineers can extend – clean APIs, config-driven behavior, well-documented evaluation • Behavioral signals from search flow back into ranking and retrieval, making the system measurably smarter over time • Ideal Background • Ideal Background • 5–8+ years of experience building and shipping search, retrieval, or ranking systems in production • Strong experience with Elasticsearch or similar search engines (Solr, Vespa, OpenSearch) – index design, query optimization, hybrid retrieval • Hands-on experience with learning-to-rank (LightGBM, XGBoost, LambdaMART) or similar applied ranking approaches • Strong Python skills and software engineering fundamentals – clean, typed, well-structured production code • Experience with embeddings and vector search – dense retrieval, ANN indexing, embedding fine-tuning • Pragmatic ML sensibility: you pick the simplest model that works, measure rigorously, and ship iteratively • Experience with offline evaluation methodology – nDCG, MRR, precision/recall at k, A/B test design and interpretation