VNG is hiring: Senior AI Engineer, GreenNode (Ho Chi Minh City)

The role

VNG's GreenNode runs RAGEngine and other AI products on its own self-hosted model platform. This role owns the AI agent side: how agents plan, call tools (function calling), keep memory and work together in multi-agent setups. You would build and tune an end-to-end agentic RAG pipeline, from ingesting and chunking documents, embedding, hybrid retrieval and reranking to query rewriting and iterative retrieval. Another big part is measuring quality: building benchmarks and automated evaluation (LLM-as-a-judge, RAGAS), then trading quality against latency and cost with semantic caching, model routing and context compression. The results go to production together with the product and infrastructure teams.

What they ask for

• 2 years (Middle) to 4+ years (Senior) of AI/ML engineering, with at least 1–2 years building LLM applications directly
• Strong Python; has used agent/RAG frameworks such as LangChain/LangGraph, LlamaIndex, AutoGen or CrewAI, or built agents from scratch
• Practical LLM know-how: prompt engineering, function calling, structured output, managing the context window
• Embeddings, vector databases (Milvus, Qdrant, Weaviate, pgvector) and search techniques such as BM25, hybrid search and cross-encoder rerankers
• Has shipped LLM systems to production: API design, monitoring, tracing (Langfuse, OpenTelemetry); reads English documentation and papers well

Good to know

• Location: Ho Chi Minh City, in Vietnam.
• Job ID 26-ENG-4070. The page shows no posting date; the job has been on VNG's careers site since 4 November 2026 and was still open on 9 November 2026.
• The title says Senior, but the requirements also accept Middle-level candidates (2+ years).

How the interview usually goes

At VNG, a CV screen is usually followed by one or two technical interviews with the team, sometimes with a take-home test, then a conversation with the manager and HR. For an AI role like this, expect deep questions on the LLM/RAG systems you have built: why you chose that design, how you measured quality, and how you handled latency and cost.