Generative AI Engineer

You will be responsible for developing, fine-tuning, and evaluating vision-language models such as CLIP, Flamingo, BLIP, GPT-4V, LLaVA, among others. Your role will involve designing and constructing multimodal pipelines that fuse image/video inputs with natural language comprehension or generation. Working with extensive image-text datasets like LAION, COCO, Visual Genome for training and validation will be a key part of your job. You will also be implementing zero-shot/few-shot multimodal inference, retrieval, captioning, VQA (Visual Question Answering), grounding, etc. It is essential to collaborate closely with product teams, ML engineers, and data scientists to deliver practical multimodal applications. Additionally, optimizing model inference performance and resource utilization in production environments using ONNX, TensorRT, etc., will be expected. Conducting error analysis, ablation studies, and suggesting enhancements in visual-language alignment, as well as contributing to research papers, documentation, or patents if in a research-oriented team, are also part of the responsibilities. To excel in this role, you should hold a Bachelors/Masters/PhD in Computer Science, AI, Machine Learning, or a related field. You must possess at least 2+ years of experience in computer vision or NLP, with a minimum of 1+ year specifically in multimodal ML or VLMs. Strong proficiency in Python programming, along with hands-on experience in libraries like PyTorch, HuggingFace Transformers, OpenCV, torchvision, is required. Familiarity with VLM architectures such as CLIP, BLIP, Flamingo, LLaVA, Kosmos, GPT-4V, etc., is expected. You should also have prior experience in dataset curation, image-caption pair processing, and image-text embedding strategies. A solid grasp of transformers, cross-attention mechanisms, and contrastive learning will be advantageous for this role. This is a full-time position with a day shift schedule. The work location is in-person.,