Prompt & Context Engineering (Core)

• Design, build, and continuously iterate production prompt systems: system prompts, few-shot design, structured output enforcement.
• Build agentic workflows: function calling / tool use, multi-step planning, error recovery, and AI harness engineering beyond simple API integration.
• Establish prompt versioning, A/B testing, and regression test suites so prompt changes are measurable rather than guesswork.
• Design and tune RAG pipelines: chunking strategy, embedding selection, retrieval tuning, and reranking.

 

Model Fine-tuning & Customization

• Build training datasets (collection, cleaning, labeling, synthetic generation) and run fine-tuning experiments on open models, with team support on infrastructure.
• Benchmark and select models across commercial APIs and open-weight models based on quality, latency, and cost.

 

Evaluation & Quality

• Help build our evaluation practice from the ground up: golden datasets, automated scoring, LLM-as-judge
• Define per-feature quality metrics, prevent regressions across model or prompt changes, and analyze failures such as hallucination, format breakage, and refusal.

 

Integration & Operations

• Expose AI capabilities to product teams through clean, well-documented service interfaces.
• Instrument observability for token cost, latency, and full request tracing.
• Define request/response contracts for AI features with front-end and back-end engineers: streaming, failure states, and long-running jobs.

Required

• 2+ years in software engineering, including at least 1 year building and shipping LLM-based features to production.
• Hands-on production experience with at least one major LLM API (Anthropic, OpenAI, or Google) and at least one open-weight model family (Llama, Qwen, Mistral, or similar).
• Hands-on fine-tuning experience (SFT, LoRA or QLoRA)
• Strong Python, with practical use of the Hugging Face stack or PyTorch.
• Experience designing prompts as engineered artifacts.
• Sufficient back-end ability to ship your own work: Python or Node.js, REST APIs, and endpoints consumed by web clients (streaming, cancellation, timeout handling).

 

Preferred

• Experience with inference optimization (vLLM, TensorRT-LLM, quantization) or GPU training infrastructure.
• Working knowledge of vector databases and embedding-based retrieval.
• Experience with multi-agent frameworks and orchestration.
• Experience with Vietnamese- or Korean-language model work, or multilingual evaluation.

 

Soft Skills

• Comfortable working with ambiguity - this field changes monthly.
• Evidence-driven: you reach for an eval set before you reach for an opinion.
• Positive, proactive, and highly collaborative.
• Comfortable reading recent articles and models, and translating them into something shippable.
• Strong analytical and problem-solving skills.

 

Application

• CV and a link to your GitHub or Hugging Face profile.
• A short description of one LLM feature or model you built: what the problem was, what you tried, how you measured whether it worked, and what you would do differently now. Half a page is enough.