Senior agentic AI engineers in Vietnam, dedicated full-time

What an agentic AI engineer is

An agentic AI engineer builds software in which a language model does more than answer: it reads your data, decides which tool to call, acts in your systems and hands over to a person when it should. The hard part is not the prompt. It is the tool design, the retrieval, the checks before an action, the evaluation and the cost control that make the agent safe to run in production.

This is a different job from coding with an AI assistant. Our page on Claude Code is about the tool our developers use every day; this page is about what these engineers build for you.

What they build

• Tool-calling agents: agents that act in your product through your own APIs, from a Claude agent with 50+ tools that publishes and replies on 5 social networks to research and translation agents on LangGraph.
• MCP servers: Model Context Protocol integrations that expose your data and actions to an agent in a controlled way.
• RAG with evaluation: retrieval over your documents with pgvector or Weaviate, answers with verbatim citations, and an evaluation harness (or RAGAS and Giskard) to catch quality regressions before your users do.
• LLM integrations: chatbots, document assistants and support tools on the Anthropic Claude, OpenAI and OpenRouter APIs, with LangChain where it helps, connected to your CRM or back office.
• Production work: streaming, prompt caching, idempotent pipelines, usage quotas and per-tenant token and cost metering, on Python (FastAPI) or TypeScript (Node.js, NestJS) back ends.

Guardrails we expect in production

An agent that can act in your systems needs the same discipline as any other production code, and a few rules of its own. These are the guardrails the engineers below have already built, and the ones we ask for on every agent project.

• Human review: a confirmation step before every irreversible action, and a handover to a person when the agent is not confident.
• Action logging: every tool call recorded, so you can see what the agent did, when and why.
• Limited access: the agent only gets the tools, data and permissions its task needs, and answers only from approved sources.
• Evaluation: a test set run on every change to prompts, tools or models, alongside the usual regression tests.
• Ownership: the developer who commits the code answers for it, reviewed through your usual process, whether a person or an AI wrote it.

Published CVs with agentic AI work

These CVs are public and describe the projects behind each capability above. Links are in the reading list at the bottom of the page. More profiles are sent with each request, within 48 hours.

• Michael P., Senior AI Engineer · LLM Agents (Claude) & Document Intelligence: Claude agent with 50+ tools and a confirmation step before every irreversible action, MCP, pgvector retrieval with verbatim citations, an evaluation harness, and review of AI-generated code as engineering lead.
• Arichus, Senior Python Developer: research and translation agents with LangGraph, a local RAG system evaluated with RAGAS and Giskard, and data platforms on Python and Airflow.
• Leon, AI Fullstack Developer: LangChain agents for customer support and document summarisation, RAG pipelines with Hugging Face embeddings and Weaviate, and model evaluation metrics.
• David, AI Engineer: a customer support system on the OpenAI API with LangChain memory, a vector database and connections to business CRMs.

How it works

• Scoping: you describe the agent or the LLM feature, your stack and your data to the founder, in French or English.
• CVs: 2 to 3 pre-selected senior profiles within 48 hours, with the rate for each profile.
• Your choice: you interview the engineers and pick the one who joins your team.
• Start: the engineer starts within 7 business days on average, full-time, 8 hours a day, 5 days a week, on your project only, with no minimum term.

Frequently asked questions

Do your engineers build AI agents, or do they only use Claude Code?

Both, and they are different skills. All our developers use Claude Code to write code. The engineers on this page also build agents into your product: tool-calling agents, MCP servers, RAG with evaluation and LLM integrations, as described in their published CVs.

Which models and frameworks do they work with?

Our published CVs cover Anthropic Claude, the OpenAI API (including the Assistants API), OpenRouter, LangChain, LangGraph, Hugging Face, MCP, pgvector, Weaviate and Ollama, on Python (FastAPI) and TypeScript (Node.js, NestJS) back ends. Tell us your stack and we send the profiles that match it.

Who owns the code and the prompts?

You do. Source code and intellectual property are 100% yours, under NDA. Prompts, tool definitions and evaluation sets written for your project are part of that code and live in your repository.

Who is responsible when an agent makes a mistake?

The developer who commits the code answers for it, reviewed through your usual process. That is why we only place senior engineers, and why we ask for human review, action logging, limited access and evaluation on every agent that acts in your systems.

How much does a senior agentic AI engineer cost?

The rate depends on the profile, its stack and the length of the engagement. It is sent with the CVs, within 48 hours. There is no recruitment fee and no minimum term.

Related reading

• Where to hire developers who use Claude Code or Cursor: options compared
• CV: Michael P., Senior AI Engineer · LLM Agents (Claude) & Document Intelligence
• CV: Arichus, Senior Python Developer
• CV: Leon, AI Fullstack Developer
• CV: David, AI Engineer
• Senior offshore teams that work with Claude Code
• Hire dedicated senior developers in Vietnam
• AI agents: the fundamentals you need for your tech projects
• What nobody tells you about AI developers in Vietnam

Get started

Tell us your stack, your team and what you need delivered. We send 2 to 3 pre-selected senior profiles within 48 hours, with the rate. No commitment.