Responsibilities
- Design and optimise hybrid lexical–semantic retrieval pipelines (e.g., BM25, dense vectors, HNSW/LSH, generative retrieval) to improve precision and recall across GoFood and GoPay surfaces.
- Build high-quality embeddings and relevance signals that capture user intent, cuisine and dish semantics, geolocation, delivery constraints, price sensitivity, and promotions.
- Develop multi-task deep ranking models that balance conversion, diversity, merchant quality, and long-term user retention, integrating real-time signals such as promotions, surge, and stock availability.
- Build personalised ranking layers and user behaviour models leveraging historical orders, preferences, and contextual features.
- Engineer recommendation algorithms using collaborative filtering, graph-based methods, and sequence models for retrieval expansion (e.g., Q2Q2I, Q2I2I, U2I), including for cold-start merchants and new dishes.
- Advance embedding quality for multi-modal data (text, images, behavioural signals) and use LLMs to enhance structured knowledge (taxonomy tagging, dish attributes, dietary labels).
- Incorporate structured metadata, taxonomy signals, and knowledge-graph features into retrieval and ranking pipelines to improve semantic understanding and consistency.