Hire Murat C. - LLM Engineer - Remote

Developed an interpretability framework at Goodfire AI for analyzing and interpreting large language models (LLMs), enhancing model transparency and explainability. The framework optimizes LLM throughput with the sparse autoencoders (SAEs) while extracting and interpreting LLM features and outputs using the vLLM backend. Co-authored the research papers Mapping the Latent Space of Llama 3.3 70B, and Under the Hood of a Reasoning Model on interpreting the behavior of Llama 3 and DeepSeek LLMs with SAEs. Co-authored the research paper Endogenous Resistance to Activation Steering in Language Models Emerges at Scale, where we showed LLMs can detect and resist manipulations to their internal activations. Worked on AI alignment research, and built image & video generation tools leveraging diffusion models.

* Developed deep reinforcement learning and control algorithms for supply chain automation and warehouse management and incorporating the designed algorithms into warehouse management systems. * Developed MLOps pipelines for the reinforcement learning algorithms in Google Cloud Platform. * Developed tools that utilizes large language models (LLMs) for searching content, document and code summarization, and interacting with databases in GCP. * Mentoring researchers for research and algorithm development in multi-shuttle and autonomous ground vehicle systems.