Masterarbeit: Wissen statt Cloud? Unternehmenswissen und Routing für lokale Sprachmodelle in KI-Coding-Agenten (all gender)
Core
Develop a gateway to route coding tasks between local LLMs and cloud models, leveraging internal corporate knowledge to minimize token usage while maintaining performance.
Role type
Research Master's Thesis (Design Science Research)
Builds
An inference gateway with a Chat-Completions-compatible interface for routing coding agents
Domain
AI/LLM engineering, Software Architecture, Empirical Software Engineering
Deliverable
production ML models
Required skills
Python, REST APIs, HTTP, Linux, Containers, LLMs, Model Routing, Embeddings, RAG, Coding Agents, Experimental Design, Statistics
Preferred skills
GPU Server operation, LLM Inference Server operation, Streaming (SSE)
Responsibilities
Research LLM routing and agent systems, implement a prototype inference gateway, select coding agents based on criteria, implement rule-based routing policy, integrate internal corporate knowledge, build an automated test catalog and evaluation harness, compare baselines statistically, replicate findings with a second agent, derive design recommendations