Backend development & delivery for RAG (incl. Graph RAG & agentic search): Contribute to building RAG backend capabilities, including retrieval, re-ranking, context construction, prompt orchestration, and response aggregation. Exposure to Graph RAG patterns (e.g., entity/relation extraction, basic graph traversal) and agentic search concepts (e.g., multi-step retrieval, tool/function calling) is a plus.
Build pipelines for multi-modal and multi-type data (documents, emails, wikis, images, tables, logs, structured sources), including extraction/OCR, chunking, enrichment, embedding, and incremental re-indexing with rollback. Retrieval, ranking & graph reasoning optimisation: Implement hybrid retrieval (BM25/vector/structured filters), multi-stage recall, reranking (cross-encoder/LLM), deduplication/diversity controls, and graph-based retrieval strategies to improve relevance, explainability, and traceability. Grounded, supported answer generation: Build generation workflows that produce grounded answers with explicit support (citations, quoted evidence snippets, and source metadata), including answer structuring, confidence/scoring signals, refusal/insufficient-evidence handling, and traceable “why this answer” outputs for auditability. Security & compliance: Implement and maintain security controls such as access control (RBAC/ABAC), data classification/masking, audit logging, encryption, and secrets management. Work with senior engineers and compliance partners to ensure retrieval and generation flows follow bank-grade requirements. Data governance & lineage: Support data ownership, classification, retention, and access policies in day-to-day development. Help maintain metadata and basic lineage/provenance (e.g., citations/attribution) so outputs are grounded in approved sources. Quality evaluation & continuous improvement: Establish offline/online evaluation (groundedness, faithfulness, citation coverage, latency), enabling A/B testing and regression validation. Cross-team collaboration: Partner with AI Infra, data platforms, information security, and product teams to land RAG capabilities in multiple systems and continuously evolve the platform.