Overview

Senior AI/ML Engineer

Responsibilities

• Design and own the clickstream and impression-logging pipeline, including the
• Normalise the catalogue and match products across gift cards, merchandise, and
• Own search quality end to end: zero-result rate, query understanding, synonym and
• Build semantic search and eligibility Q&A that answers natural-language questions
• Build co-purchase and cohort-based collections from order history, such as
• Build session-based recommendations and in-session intent modelling, such as
• Build personalized collections from browse and redemption behaviour
• Build a learned ranker over user, item, cohort, and context features, with eligibility,
• Build utilization intelligence that identifies who is at risk of forfeiting funds, what
• Enforce a hard architectural boundary around health-adjacent interaction data:
• Treat confidently wrong eligibility answers as the worst failure mode, and design
• Deploy, serve, and operate models in production, owning latency, availability, and
• Build offline evaluation that reflects online outcomes instead of eyeballed results
• Run A/B tests on utilization rate, order frequency, and search success, the metrics
• Own and report the business outcomes of your work, connecting model and system
• 5+ years in production ML or ML-adjacent engineering
• Depth in at least one of search and ranking, or recommendations and
• Shipped a production system from zero that you can walk us through in detail,
• Strong on embeddings and retrieval: vector search (FAISS, HNSW, pgvector, or a
• Comfortable with learned ranking, including GBDT rankers and neural or sequential
• Expert Python, strong SQL, and willing to own pipelines end to end (Spark or

Qualifications

• Locations: Delhi
• 5 Days working
• Rigorous about evaluation: you build offline eval sets instead of eyeballing outputs,
• Evidence of self-direction: you have set your own roadmap, negotiated scope with
• Marketplace, insurance, healthcare, fintech, or loyalty experience, anywhere the
• Clickstream instrumentation designed from scratch
• Having been the first ML hire somewhere

Nice to have

• RAG or grounded-generation systems in production, with a real strategy for
• Constrained ranking, where eligibility, budget, inventory, or regulatory filters sit over
• Privacy-preserving or privacy-constrained ML, where data segregation or regulatory
• Entity resolution, product matching, or taxonomy work at scale