AI Engineer

Major projects you will tackle

• Omni-channel customer chatbot that handles FAQs, after-sales, and ticket booking for bus, flight, train, and vehicle-rental verticals—across text and voice.
• Department-specific assistants for HR onboarding, finance invoice queries, operations incident triage, IT help-desk, and more—each powered by shared LLM components and n8n triggers.
• Company-wide automation hub built on n8n, including reusable nodes and guard-rails that let non-technical teams create flows safely.
• Multimodal expansion that blends text, speech, and images so customers can, for example, upload a ticket photo or speak a booking change request.
• You will fine-tune foundation models, fuse retrieval with LLM reasoning, and iterate in Vietnamese, English, and other languages.
  

What you will do

• Design, implement, and continually improve our multi-agent framework. Build and refine agent components such as short- and long-term memory stores, planning/reflection loops, agent-to-agent messaging, and specialised prompt templates that let multiple agents collaborate on complex tasks.
• Select and adapt OpenAI, Gemini models, Llama, Mixtral or better—using LoRA or full fine-tuning—so the models speak our brand voice in multiple languages.
• Build and optimise retrieval-augmented pipelines, keeping latency below two seconds and hallucinations under five percent.
• Craft prompts, refusal rules, and jailbreak tests; automate factuality, safety, and multimodal hallucination checks in CI.
• Package models with Docker and serve them through FastAPI or gRPC behind vLLM, Triton Inference Server, or Text-Generation-Inference; add monitoring with Grafana / Prometheus for latency, drift, and GPU cost (for later milestones).
• Pair daily with Conversation Designers, MLOps engineers, Automation engineers, and business stakeholders; publish model cards, data sheets, and rollback plans.

Must-have qualifications

• 1+ years of experience shipping NLP or generative-AI systems to production.
  Strong Python and PyTorch skills plus heavy use of Hugging Face (Transformers, PEFT, Datasets).
• Practical knowledge of vector databases such as milvus, pgvector, Qdrant, or Pinecone and hybrid-search techniques.
• Demonstrated ability to raise containment or reduce hallucinations through data-driven experiments.
• Solid engineering habits: Git, code reviews, unit and integration tests, CI/CD pipelines.
• Clear spoken and written English; Vietnamese fluency is a bonus.
  

Nice-to-haves

• Hands-on experience with Rasa, Dialogflow CX, or ASR/TTS pipelines for voice bots.
• Deep GPU-performance tuning, including quantisation, KV-cache optimisation, or custom Triton kernels.
• Familiarity with multi-agent frameworks such as PydanticAI, CrewAI, LangGraph or others.
• A privacy-by-design mindset and working knowledge of GDPR or PDPA compliance.