LEAD SOFTWARE ENGINEER

We’re hiring an AI Engineer to design and build production-grade GenAI and ML services on GCP. You’ll own core platform components for hybrid RAG, vector + graph data stores, multi-agent orchestration, and secure tool connectivity (including MCP-style patterns). This is a purely technical role with strong ownership from design through running. Key responsibilities Build backend GenAI/ML services (GCP-first)  Design and implement scalable API-first services for LLM applications (chat/search/assistant capabilities).  Build low-latency, high-throughput systems with robust caching, batching, retries, idempotency, and fallbacks.  Integrate with GCP services (e.g., Cloud Run/GKE, Pub/Sub, Cloud Storage, Secret Manager, Cloud Logging/Monitoring). Multi-agent orchestration (ADK-style frameworks)  Implement agent orchestration patterns: planner–executor, tool-routing, specialist agents, critic/reviewer loops.  Build state management, memory, and workflow execution with clear traceability and deterministic behavior where needed.  Create reusable agent/tool SDK components for other engineering teams. Hybrid RAG pipeline (dense + sparse)  Build a hybrid retrieval layer (embeddings + keyword/BM25), reranking, metadata filters, query rewriting, and context compression.  Implement ingestion pipelines: chunking strategies, deduplication, metadata enrichment, and incremental re-indexing.  Ensure response grounding with citations, provenance, and strict access controls. Vector DB + Graph DB platform components INTERNAL  Own vector indexing strategy, tenant isolation, lifecycle/retention policies, and performance tuning.  Build graph-backed retrieval (GraphRAG / relationship-aware search) using entity linking and multi-hop traversal.  Design data models and services that combine vector similarity + graph context for better precision/recall. MCP-style tool connectivity & enterprise integration  Implement secure, standardised tool connectors (MCP-style) to internal APIs/data sources.  Enforce authentication/authorisation, rate limits, audit logs, and policy checks per tool invocation.  Provide a registry/catalogue of tools with versioned schemas and compatibility guarantees. LLMOps/MLOps: reliability, evaluation, observability  Build automated evaluation harnesses (golden sets, regression tests, red- teaming, retrieval metrics).