AI Software Engineer Intern

SemiAI is a deep-tech AI company building software for semiconductor manufacturing. The AI Software Engineer Intern will build, evaluate, and improve LLM-powered agents and machine learning systems using real-world semiconductor process data, while turning prototypes into usable tools and presenting weekly demos.

What you'll do:

• Build agents and LLM-powered tools — tool calling, retrieval, planning loops, structured outputs
• Design the evaluations that tell you whether the system actually works, not just whether it ran
• Explore and clean real customer data: undocumented columns, missing values, sensor noise
• Build and test models on process data — and be honest about what the numbers mean
• Turn prototypes into things your teammates can run, with documentation that holds up
• Present a weekly demo: what you built, what broke, what you learned

What they're looking for:

• Current student or recent graduate in Computer Science, Data Science, Engineering, Statistics, Physics or a related field
• Strong Python — comfortable with pandas and NumPy, and comfortable reading code you didn't write
• Hands-on experience building with LLMs: API calls, prompting, tool use, retrieval, or agent loops. Personal projects count fully
• Solid ML fundamentals — train/validation/test discipline, evaluation metrics, and an instinct for when a result looks too good to be true
• Fluent with AI coding tools such as Claude Code, Cursor or Copilot, and thoughtful about where you trust them and where you don't
• Real curiosity about a physical, industrial domain you've never worked in
• Authorized to work in the United States
• You've built something agentic end to end — a multi-step agent, an MCP server, a RAG pipeline, an eval harness
• Experience with agent or LLM frameworks: LangGraph, LlamaIndex, the OpenAI or Anthropic SDKs, MCP
• Experience with evaluation and observability tooling — LangSmith, Weights & Biases, Braintrust, or something you wrote yourself
• Time-series or sensor data experience
• Vector databases, embeddings, fine-tuning