Intern AI Application Engineer (RAG & Agentic Systems)

🔑 Key Responsibilities

• Design and implement production-grade Advanced RAG pipelines utilizing Hybrid Search, Re-ranking, Query Expansion, and Multi-vector retrieval.
• Architect structured document ingestion pipelines, mastering advanced chunking strategies and complex metadata filtering schemes.
• Deploy, serve, and optimize state-of-the-art open-source foundational models and Small Language Models (SLMs) such as Qwen, DeepSeek (including V4 architectures), and Codestral.
• Maximize throughput and minimize time-to-first-token (TTFT) on limited GPU infrastructure using advanced serving engines.
• Build context-aware, reliable AI agents capable of autonomous planning, complex tool reasoning, and task execution.
• Leverage advanced multi-agent orchestrations to handle cyclic graph execution and persistent state loops.
• Establish rigorous automated AI quality gates to evaluate hallucination rates, retrieval accuracy, and generation fluency.
• Implement end-to-end trace logging and monitoring systems to flag search degradation and production system failures in real time.

✅ Requirements

• Currently in your 4th year of an IT/Computer Science degree or a recent graduate.
• Experience managing specialized vector databases (Qdrant, Milvus) alongside mastery of semantic and recursive text splitting.
• Comprehensive understanding of advanced search mechanisms, including Cross-Encoders vs. Bi-Encoders, Reciprocal Rank Fusion (RRF), and Self-RAG architectures.
• Production-proven proficiency with LangGraph or LlamaIndex Workflows.
• Practical expertise configuring, benchmarking, and maintaining model serving layers through vLLM or llama.cpp (including GGUF/AWQ quantization methods).
• Strong technical English communication skills to document architectural patterns and collaborate with senior engineering stakeholders.

🌟 Nice-to-Have

• Experience constructing and integrating Knowledge Graphs into standard vector search pipelines to preserve complex relational context.
• Basic understanding of CUDA/Triton; the ability to write or customize custom kernels to accelerate inference loops is a significant advantage.
• Deep understanding of handling error-prone AI systems using enterprise patterns like Dead Letter Queues (DLQ), Circuit Breakers, and exponential backoff retry logic.
• Experience with Parameter-Efficient Fine-Tuning (LoRA / QLoRA) for small model modules targeting niche downstream tasks.

WHY YOU LOVE WORKING HERE

• Attractive Intern salary
• Motorbike or bus support allowance.
• Have a chance to be official member of ISB Vietnam.
• Professional and challenge working environment.
• Japanese culture and working atmosphere discovery.