Description
I am building a no-code platform that enables non-technical users to convert small language models into multimodal vision-language models. I already have a working Python-based training and inference pipeline that combines a language model with a vision encoder using PyTorch, PyTorch Lightning, Hugging Face, CLIP, LoRA, Gradio, MLflow, DVC, Docker, and GitHub Actions. I need an experienced full-stack AI/ML engineer to help turn this existing pipeline into a user-friendly web product. The MVP should allow users to: Select a supported language model and vision encoder Upload or select a training dataset Configure basic training parameters Start and stop GPU training jobs Monitor training progress, logs, losses, and errors Save and manage model checkpoints Test the trained model through an image-and-text inference interface Deploy or export the resulting model Preferred technology stack: React or Next.js frontend Python and FastAPI backend PyTorch, Hugging Face Transformers, and PyTorch Lightning Docker and cloud GPU infrastructure WebSockets or similar technology for live job updates Experience with model training, inference, queues, storage, and MLOps The first milestone will support one model architecture and one GPU provider. Please include relevant AI platform, MLOps, or model-training projects in your proposal.