Recent work, in production.

• Fine-tuned vision-language and diffusion models for applied product use cases
• Sped up training with mixed precision and parallel evaluation
• Adapted models cheaply with LoRA, QLoRA and DreamBooth
• Improved multimodal retrieval with CLIP-style embeddings

PyTorch, Flux-Dev, CLIP, Diffusers, LoRA, QLoRA, DreamBooth

Case study →

• Built deduction classification models with XGBoost and LightGBM
• Raised accuracy through targeted feature engineering
• Automated monitoring to catch drift before it caused failures

XGBoost, LightGBM, Python, Feature engineering, Drift monitoring

Case study →