Overview

Work on generative models for visualization tasks.

Responsibilities

• Design, train, and fine-tune large-scale diffusion models (DDPM, DDIM, LDM, DiT) for image, video, and multimodal generation tasks.
• Drive improvements in sampling efficiency distillation, consistency models, progressive training, and guided generation techniques.
• Stay current with and rapidly prototype ideas emerging from the broader AI community. Computer Vision & Perception
• Build production-grade pipelines for image/video understanding: segmentation, detection, depth estimation, optical flow, and 3D reconstruction.
• Develop and fine-tune vision foundation models (ViT, CLIP, DINOv2, SAM) for downstream tasks using parameter-efficient methods (LoRA, adapters).
• Integrate vision encoders with generative backbones for controllable generation (ControlNet, IP-Adapter, inpainting, editing). Applied ML & Systems Own the full ML lifecycle: data curation, experiment tracking, model evaluation, optimization, deployment, and monitoring.
• Optimize models for inference: quantization (INT8/FP8), ONNX export, Flash Attention, and xFormers.
• Design scalable training infrastructure on distributed GPU clusters (DDP, FSDP, DeepSpeed) across thousands of GPU-hours.
• Define and instrument evaluation frameworks, benchmarks, and human preference studies (RLHF / DPO) to measure generative quality. Leadership & Collaboration
• Lead technical design reviews, write engineering RFCs, and set quality standards for the team. Mentor junior and mid-level ML engineers through code reviews, 1:1s, and pair-programming sessions.

Qualifications

• 10 - 14 years of hands-on ML engineering
• experience in industry or research. MS or PhD in Computer Science, Machine Learning, Statistics, or equivalent practical experience.
• Expert-level Python; strong, mandatory proficiency in PyTorch . Deep theoretical and practical knowledge of score-based and diffusion models . Strong background in computer vision fundamentals: CNNs, ViTs, feature pyramids, multi-scale processing. Experience fine-tuning large vision and generative models at scale. Proficiency with distributed training frameworks (DDP, FSDP, DeepSpeed, Megatron-LM). Solid grasp of probabilistic ML, variational inference, and information theory.
• Experience with MLOps tooling (Weights & Biases, MLflow, DVC, or equivalent). Track record of shipping ML models to production at scale. Excellent written and verbal communication skills with cross-functional stakeholders. Preferred Qualifications
• Experience with flow-based generative models (normalizing flows, CNFs, Rectified Flow, Flow Matching).
• Experience with video generation models (Sora-style architectures, CogVideo, AnimateDiff, SVD).
• Familiarity with 3D generative models (NeRF, 3D Gaussian Splatting, Zero-1-to-3, Point-E). Background in multimodal systems (LLMs + vision: GPT-4V, LLaVA, InstructBLIP-style architectures).

Education

• MS or PhD in Computer Science or a related field.