Agentic AI Engineer
Budapest
Job No. r00311970
Full-time - Hybrid
工作描述
At Accenture Industry X we are building an AI Hub to design and scale agentic AI solutions—multi‑agent systems that can reason, plan, and take action by combining LLMs, retrieval‑augmented generation (RAG), and external tools/APIs.
In this role, you will develop the core building blocks for reliable agent systems: orchestration patterns, agent‑to‑agent collaboration, state and memory management, and production‑minded testing/observability. The goal is to move from prototypes to repeatable, robust, and maintainable agentic workflows.
Responsibilities
- Design and implement agent‑to‑agent communication patterns and multi‑agent collaboration workflows.
- Integrate LLMs, RAG pipelines, and external APIs/tools into end‑to‑end agent systems.
- Develop, test, and iterate on orchestration logic (routing, planning, tool selection, error handling, and structured execution).
- Build prompt templates, memory/state management, and reusable workflow components to improve reliability and consistency.
- Implement and optimize RAG + embedding workflows (retrieval strategy, chunking, evaluation, iteration).
- Own testing & debugging for multi‑agent systems, including tracing, evaluation, and regression testing across workflows.
职位要求
Required skills
Please submit a CV that highlights your proven experience, with detailed descriptions of relevant home or personal projects
- Hands-on experience building LLM-based applications in Python, including structured prompting and multi-step workflows.
- Practical experience with at least one agent framework or orchestration approach (e.g., LangChain or a comparable framework). Experience with LangGraph is a plus.
- Working knowledge of prompt templates and basic memory/state concepts (e.g., session memory, conversation state, lightweight persistence).
- RAG fundamentals with some implementation exposure: ability to build a basic retrieval flow (data → chunking → embeddings → retrieval → grounded answer) and iterate on quality.
- Debugging mindset for agentic workflows: comfortable troubleshooting tool-calls, retrieval misses, and prompt/chain issues; familiar with logs/traces or basic observability practices.
- Strong English language knowledge.
Nice to have
- LangGraph experience (graph/state-machine orchestration) and/or multi-agent patterns.
- Langfuse (or similar tracing/evaluation tooling) for prompt/version tracking, traces, and quality monitoring.
- Experience improving RAG quality beyond the basics (evaluation, retrieval strategies, chunking experiments).
- Exposure to testing practices for LLM apps (lightweight evaluation, regression checks, reproducibility habits).
