Software Engineering Intern (AI Agents & Local LLMs) — Private Equity (Hamburg / Hybrid)

Software Engineering Intern (AI Agents & Local LLMs) — Private Equity (Hamburg / Hybrid)

Location: Hamburg (hybrid preferred)

Duration: 3–6 months (internship)

Start: ASAP / flexible

Compensation: Paid internship

Language: Fluent German and English needed

What you’ll build

You’ll help us design and ship AI agents that meaningfully accelerate a direct private equity investment team—from sourcing to diligence to portfolio support. The work is hands-on: you’ll build production-grade workflows (not demos) using Claude Code for agent development and a local LLM setup for privacy-sensitive tasks.

Core projects (examples)

• Claude Code agent workflows for:
• Deal/company research and structured briefs (with citations)
• Diligence support: extracting key facts from PDFs/data rooms and generating checklists & memos
• Portfolio ops assistants: drafting outreach, summarizing calls, creating action trackers
• Local LLM stack (privacy-first):
• Set up and evaluate a local model pipeline (e.g., quantized inference, GPU/CPU options)
• Implement RAG over internal documents with access controls
• Build evaluation harnesses for accuracy, hallucinations, and cost/latency tradeoffs
• Tooling & reliability
• Guardrails, structured outputs (JSON schemas), logging/tracing, basic tests
• Prompt/tool versioning and regression checks

Tech stack (what we use / want you to learn)

• Claude Code for agent development and orchestration
• Python (preferred) and/or TypeScript
• APIs + tooling for LLMs, structured tool calling, and RAG
• Vector search (pgvector / similar), Postgres, Docker
• Optional: local inference frameworks (e.g., vLLM / llama.cpp / Ollama-style runtimes), evaluation tooling

(You don’t need all of this on day 1—strong fundamentals + willingness to learn fast matters more.)

What we’re looking for

Must-have

• Strong CS fundamentals and practical coding ability (Python or TS)
• Evidence you ship: GitHub/projects, hackathon work, or a small demo you can walk us through
• Comfort working with messy real-world data (PDFs, tables, websites)
• Clear communication and ownership: you can drive a task from “idea → working prototype → improved v2”
• Deep Knowledge of Claude Code and the entire AI Agent development

Nice-to-have

• Prior experience with LLM agents, RAG, vector DBs, prompt/tool calling
• Interest in privacy/security and “local-first” ML systems
• Familiarity with evaluation: test sets, automated checks, error analysis

By the end of the internship you will have shipped:

• At least 4–5 agent workflows used by the investment team weekly
• A local LLM prototype integrated into one workflow (with measured quality/latency)
• An evaluation and monitoring loop so the system improves over time

Why this role is different

• You’ll build real internal systems used by decision-makers
• You’ll learn how to make agents reliable (not just impressive)
• You’ll work on privacy-first AI for a high-context domain (direct PE)

How to apply

Send:

1. A short note: one project you built and what you owned
2. An agent you have developed (even a weekend prototype)