Worked on the meilisearch/meilisearch repository to expand transformer model compatibility and improve model loading reliability. Developed ModernBert support within the Hugging Face embedder, enabling seamless embedding for both Bert and ModernBert architectures by updating the embedder’s loading and processing logic. Addressed model weight compatibility by implementing safetensors tensor name normalization, standardizing tensor names for models lacking the 'model.' prefix and ensuring consistent inference behavior. Leveraged Rust for robust data processing and integrated machine learning and natural language processing techniques to enhance embedding flexibility. The work focused on practical improvements to model support and reliability within the codebase.