Stellenbeschreibung:

About the Company
We specialize in creating conversational AI experiences for roleplay, gaming, social interactions, and creative writing.


As our LLM Engineer, you'll fine-tune and optimize large language models that power conversations for over 30 million users, processing more than 5 million messages daily. You'll be at the forefront of developing AI companionship technology that scales globally while maintaining personalized and meaningful interactions.


Key Responsibilities
• Interact with stakeholders (Co-founders, Web Engineers, DevOps Engineers) to bring your project to life.
• Oversee the creation and optimization of algorithms for LLM behavior adjustments based on user interactions, focusing on fine-tuning and prompt engineering.
• Develop features to improve the richness of the product (multi-character chats, gamification, etc)
• In addition to chat, interacting with modalities managed by other team members (audio, image, video), and collaborating with them
• Adaptation and fine-tuning of base models for multilingual support
• Manage the creation and maintenance of diverse datasets critical for training and improving the performance of LLMs.
• Assess and determine the best technological approaches, selecting between classifiers, fine-tuning, and other methods based on the specific project's needs.


Your Qualifications
• Python Mastery: 5+ years building production-grade, modular, maintainable codebases
• LLM Architecture Expertise: Deep understanding of transformers and their training dynamics (attention, positional encodings, samplers, tokenizers, post- training, reasoning LM)
• Inference Optimization at Scale: Expert with vLLM / TensorRT-LLM (or similar); proven record of reducing latency and memory via quantization and/or distillation
• Distributed Training: Hands-on multi-GPU / multi-node fine-tuning using FSDP, DeepSpeed, or accelerate; comfortable with mixed-precision, gradient checkpointing, and memory-aware scheduling
• Performance Profiling & Optimization: Skilled at identifying and resolving compute or memory bottlenecks across CPU/GPU pipelines with industry-standard profiling workflows


Nice-to-Haves
• Concurrency & Runtime Engineering: Strong with asyncio, multiprocessing, or equivalent backend/batch-scheduling patterns
• Low-level Systems: Practical CUDA / Triton experience; able to write or debug custom kernels
• Open-Source Impact: Contributor to core LLM tooling (vLLM, HF Transformers Triton, etc.)
• Real-time Deployments: Built or maintained latency-critical, multi-user LLM services (RAG, streaming, agents, chatbots)
• Specialized Generation Use Cases: Exposure to erotic role playing, multi-turn instruction tuning, or non-English quality alignment

NOTE / HINWEIS:

EN: Please refer to Fuchsjobs for the source of your application

DE: Bitte erwähne Fuchsjobs, als Quelle Deiner Bewerbung