Working Student - LLM Quantization & Inference Performance

About Tensordyne (formerly Recogni) AI is transforming our world. It can perform cognitive functions that previously only humans could do, such as perceiving interactions across different modalities and environments - with the ability to quickly learn and then solve complex problems. Tensordyne is an AI system solution company that builds very high-performance, low-power generative AI inference systems. Our mission, through the creation of custom silicon, hardware and software, is to enable multimodal Generative AI inference acceleration at scale, with safe, sustainable, high-performance systems for our hyperscaler and neocloud data center customers. We are at the leading edge of advancing the latest research and product improvements for generative Al inference solutions that will make Al even more advantageous for compelling new generative AI applications.Tensordyne is a well funded, fast-paced startup company with headquarters in both Sunnyvale, CA, and Munich, Germany. We also have many talented team members working remotely across North America and Europe. We take care of our people and their families with comprehensive benefits, competitive compensation, flexible spending options, and recognition programs, because building category-defining technology starts with a healthy, supported team. Come join us as we shape the future of multimodal generative artificial intelligence! Role summary How much numerical precision does an LLM need, and where in the model architecture does it matter most? When does quantization improve runtime, and what does it cost in model quality? We're looking for a motivated working student to help turn these questions into experiments and experiments into maintainable code. You'll explore optimization methods and build reusable evaluation tooling. You'll work closely with our quantization, compiler, and runtime engineers, and your findings will directly influence how we optimize models for Tensordyne hardware. What you will do Model quality evaluation: Measure how low-precision computation affects performance on coding and reasoning benchmarks. Diagnose where quantization has the greatest impact. Quantization research: Implement and compare post-training quantization methods to improve inference efficiency while preserving task accuracy. Build reproducible tooling: Automate evaluations, track configurations and results, and make experiments accessible to engineers and AI agents. Measure runtime performance: Characterize latency, throughput, and memory usage to understand bottlenecks under different deployment scenarios. A first project could involve exploring the quality–performance Pareto frontier of a Qwen or LlaMa model. You'll experiment with 16-bit, 8-bit, and 4-bit quantization configurations to find the best runtime-quality tradeoff. Core qualifications Current enrollment in Computer Science, Mathematics, Electrical Engineering, Machine Learning, or a related master's degree program. Strong Python and PyTorch skills, including the ability to understand and modify model implementations. Effective use of AI coding tools and critical review of their output. An experimental mindset: you can formulate hypotheses, design comparisons, and draw careful conclusions from noisy results. Clear communication and the ability to present experimental findings and technical tradeoffs to the team. Strong pluses: experience with quantization, LLM evaluation, performance profiling, or inference optimization. This role is for you if you thrive on responsibility, learn quickly, and are naturally proactive . What we offer Join a deep-tech team building the next generation of AI inference systems. Put your university knowledge into practice with leading open-source LLMs, deepen your understanding of quantization and inference optimization, and build skills in experimental design, software engineering, and AI coding tools. You'll own a focused project, receive close