AI Agent Engineer

About Binance

Binance stands as a premier global blockchain platform powering the largest cryptocurrency exchange by both trading volume and registered users. Trusted by over 300 million people across more than 100 countries, Binance is renowned for its top-tier security measures, transparent handling of user funds, rapid trading engine, deep liquidity, and a diverse portfolio of digital asset products. Its services span trading, finance, education, research, payments, institutional offerings, and Web3 functionalities. Binance aims to create an inclusive financial ecosystem leveraging blockchain and digital assets to further monetary freedom and enhance global financial accessibility.

Role Overview

The AI Agent Engineer will join the AI Infrastructure team, positioned at the convergence of cutting-edge AI model capabilities and practical deployment of AI agents. This role blends research and engineering, requiring collaboration with both researchers and engineers to expand the limits of AI agents' functionalities. This includes advancements in Agentic Retrieval-Augmented Generation (RAG), context handling, autonomous task completion, self-evolving agents, and coordination among multiple agents. Candidates must be experienced users of agent tools with strong opinions on AI model behavior, capable of generating novel ideas and rapidly iterating prototypes based on user feedback.

Key Responsibilities

• Design and manage advanced retrieval pipelines that surpass traditional single-fetch methods, featuring adaptive, self-corrective, and multi-hop retrieval strategies. Architect Agentic RAG systems incorporating dynamic retrieval control, query decomposition, iterative retrieve-reflect-refine cycles, and collaboration among multiple retrieval agents.
• Work closely with researchers and engineers to identify and implement model-driven innovations concerning context management, long-term memory, architectures involving subagents and multiple agents, self-evolving agents, and execution of real-world tasks.
• Develop domain-specific benchmarks and evaluation methods for retrieval and agent systems. Build benchmark datasets, create annotation processes, and consistently measure and improve agent intelligence metrics such as retrieval efficiency, latency, groundedness, and task success rates.
• Utilize diverse sources of user feedback and data from production to inform research priorities. Design experiments and datasets that enhance agent and retrieval performance in live environments.

Required Experience and Skills

• At least one year of practical experience working with large language models (LLMs), retrieval-augmented generation (RAG), and AI agent systems deployed in production environments.
• Proven ability to build comprehensive retrieval pipelines, including embedding models (e.g., BGE, OpenAI), vector stores (such as Qdrant, Milvus, Pinecone, Weaviate), hybrid search methods combining keywords and vectors, and reranking models. Strong knowledge of text chunking, cleaning, multimodal data processing, and implementation of Agentic RAG patterns like Self-RAG, Corrective RAG, adaptive retrieval, multi-hop decomposition, and iterative retrieval cycles.
• Experience with Agent Harness engineering including runtimes like Pi Agent or AgentScope 2.0 or similar orchestration frameworks. Skills including session recovery, sandbox isolation, middleware/hook systems, multi-tenant support, plan and execute loops, and retrieval-grounded tool invocation.
• In-depth understanding of LLM and agent concepts such as APIs, KV caching, agent loops, tool usage, reasoning, planning, skills, memory components, subagent and multi-agent configurations. Strong proficiency in prompt and context engineering.
• Demonstrated capability to independently analyze ambiguous challenges, conceive original solutions, and drive research projects from conception to prototype with agile iteration based on experimentation.
• Frequent user and integrator of agent-based products and tools as part of daily workflows, possessing refined judgment regarding model performance and behaviors.
• Competence in fast-paced, AI-native software development leveraging AI-assisted coding across various programming languages, frameworks, and domains, along with a quick learning curve for new technologies.

Preferred Qualifications

• Extensive hands-on usage of agent platforms such as Claude Code, OpenClaw, Cowork, Manus or comparable tools already integrated into daily work or life.
• Experience with retrieval evaluation frameworks like RAGAS, TruLens, or bespoke benchmarking for retrieval quality, grounding, and latency analysis.
• Expertise in GraphRAG or knowledge graph-enhanced retrieval methods.
• Familiarity with Agent Harness systems including middleware architecture, multi-tenant session management, plugin frameworks, and sandbox environments.
• Background in model training processes, Reinforcement Learning with Human Feedback (RLHF), or combined model and system design.
• Exposure to multi-provider LLM proxies such as LiteLLM.
• Knowledge of Kubernetes/EKS for pod isolation, resource allocation, and secret management.
• Security engineering skills including prompt injection mitigation, sandbox hardening, and guardrail enforcement.

Work Environment and Benefits

• Influence the future of blockchain technology by working within the world’s foremost blockchain ecosystem.
• Collaborate with exceptional global talent within a user-focused and flat organizational structure.
• Engage in challenging, fast-moving projects with considerable autonomy in an innovative setting.
• Benefit from a results-oriented workplace offering substantial opportunities for career advancement and continuous skill development.
• Competitive remuneration and company benefits are provided.
• Work-from-home options are available depending on team needs and nature of work.

Equal Opportunity and Privacy

Binance is committed to fostering an inclusive workforce that enhances its success. Applicants confirm acknowledgment and acceptance of the candidate privacy policy. Additionally, AI tools may partially support recruitment processes for resume review and assessment, though final hiring decisions are made by humans. Further privacy information can be provided upon request.