Agentic Systems Engineer Attachee

About the role

As an Agentic Systems Engineer Attachee, you will build the multi-agent systems behind Code Particles' finance and operations stack: agents that reconcile transactions, chase payments, monitor risk and keep day-to-day operations running, all coordinated under human oversight. The role is an industrial attachment (internship) based in Nairobi.

Key responsibilities

• Design and build multi-agent workflows that automate real finance and business operations
• Connect agents to internal tools and APIs using protocols such as MCP (Model Context Protocol)
• Implement agent memory, context management and retrieval pipelines
• Build guardrails for agent actions: permissions, spending limits and human-approval checkpoints
• Optimize cost, latency and reliability of LLM-powered systems in production
• Instrument agent decisions with tracing and observability so they can be audited by humans
• Write behavioral evals and tests for agents alongside the AI Evaluation team
• Research emerging agentic patterns and share findings with the engineering team

Required profile

• Currently pursuing a degree in Computer Science, AI/ML, Software Engineering, Mathematics or a related field, requiring an industrial attachment for academic credit
• Institution must provide an attachment letter and an evaluation framework (logbook, supervisor assessment or equivalent)
• Systems thinking: comfortable reasoning about failure modes, retries and state
• Curiosity about where agentic AI is heading and its implications for finance and operations

Required skills

• Proficiency in Python or TypeScript (ideally both)
• Experience consuming and building APIs
• Working knowledge of Git and version control workflows
• Basic understanding of how large language models work (prompting, context windows, tool use)

Nice to have: projects built on the Claude, OpenAI or open-weight model APIs; experience with agent frameworks (Claude Agent SDK, LangGraph, CrewAI) or writing MCP servers; familiarity with vector databases and RAG architectures; awareness of AI safety and responsible deployment practices.