Zurück zur Liste

Stellenbeschreibung

Mid/Senior AI LLM Engineer in Deutschland

Jobgether

  • RemoteArbeiten von überall
  • VollzeitStandard-Wochenarbeitszeit

Deutschland

This position is listed on behalf of a partner company, who manages all applications and next steps. Our partner is looking for a Mid/Senior AI LLM Engineer based in Germany. As a Mid/Senior AI LLM Engineer, you will develop, fine-tune, and optimize self-hosted large language models that power personalized, engaging AI conversations. You will work across the full model development lifecycle, from preparing training datasets and running GPU-based experiments to evaluating performance and deploying improvements into production. Your work will help enhance conversational quality, long-term memory, personality consistency, and multilingual interactions. Collaborating with AI engineers, product specialists, web developers, and DevOps teams, you will turn experimental findings into reliable, scalable solutions. This hands-on role offers the opportunity to work with open-weight models, advanced fine-tuning techniques, and practical AI evaluation frameworks. You will join a remote-first environment focused on building innovative, consumer-facing AI products while continuously improving model performance and user experience. Accountabilities - Fine-tune and optimize existing open-weight large language models to improve conversational quality, response accuracy, consistency, and personalization. - Prepare, curate, and enhance training datasets to support effective model learning, reduce undesirable outputs, and improve the relevance and naturalness of generated responses. - Design and implement evaluation frameworks, benchmarks, and testing methodologies to measure model quality, identify weaknesses, and validate improvements. - Improve conversational memory, personality consistency, and the ability of models to maintain coherent, context-aware interactions over extended conversations. - Develop multilingual conversational capabilities and support interactions involving multiple characters, ensuring natural and consistent behavior across different scenarios.

- Investigate model failures and performance limitations, adjusting training data, prompts, model configurations, and response filters to address identified issues. - Conduct GPU-based training and fine-tuning experiments, troubleshoot infrastructure and performance issues, and optimize training workflows for efficiency and reliability. - Apply modern model adaptation and alignment techniques, including supervised fine-tuning, LoRA, QLoRA, and preference-based optimization methods where appropriate. - Deploy improved models into production environments, monitor their performance, analyze user-facing outcomes, and implement iterative refinements based on evaluation results. - Collaborate with AI engineering, product, web development, and DevOps teams to integrate model improvements into consumer-facing applications and align technical work with product objectives. - Work with sensitive conversational use cases and moderation requirements, balancing model behavior, user experience, and applicable content policies. Requirements: - Proven hands-on experience fine-tuning open-weight large language models and deploying models into production environments. - Strong Python programming skills, including experience developing, testing, and maintaining machine-learning workflows. - Practical experience with PyTorch and Hugging Face or comparable machine-learning frameworks and model development tools. - Experience with model training and adaptation techniques such as supervised fine-tuning (SFT), Low-Rank Adaptation (LoRA), and Quantized Low-Rank Adaptation (QLoRA). - Familiarity with preference-based training and alignment techniques, such as Direct Preference Optimization (DPO) or Group Relative Policy Optimization (GRPO). - Experience preparing and managing training datasets, improving data quality, and developing evaluation frameworks to assess model behavior and output quality.

Ähnliche Stellenangebote werden geladen…