AI Research Engineer (Model Compression & Quantization)

Tether is a global financial technology company building reserve-backed digital tokens and blockchain infrastructure that powers exchanges, wallets, and payment systems worldwide. The company operates across multiple verticals including its flagship stablecoin USDT, energy solutions, AI-driven data infrastructure, and educational platforms. This role sits within Tether's Data division, where you'll work on a fully remote team advancing the frontiers of efficient AI systems.

About Tether

Tether has established itself as a trusted player in digital finance, with USDT serving as one of the most widely adopted stablecoins globally. Beyond financial products, the company has diversified into complementary areas: Tether Power manages sustainable Bitcoin mining operations, Tether Data develops peer-to-peer communication tools and AI infrastructure, and Tether Education provides digital learning resources. The organization operates as a distributed global team, attracting talent from around the world to tackle complex problems at the intersection of blockchain, AI, and infrastructure.

What you'll do

• Apply quantization techniques including both quantization-aware training and post-training quantization to shrink model sizes and reduce inference latency for large language models, vision-language models, and multimodal systems while maintaining output quality.

• Use knowledge distillation methods to compress larger teacher models into efficient student variants capable of performing multimodal reasoning across text, image, and audio.

• Deploy pruning strategies to eliminate redundant parameters and attention heads, cutting computational requirements without degrading task performance.

• Evaluate trade-offs between model efficiency metrics (memory footprint, latency, throughput) and accuracy across different compression approaches, then propose enhancements grounded in experimental evidence.

• Investigate advanced compression strategies such as mixed-precision quantization, adaptive pruning schedules, and intermediate feature-matching distillation to optimize the efficiency-accuracy balance.

• Monitor emerging research in model compression for multimodal and generative architectures to keep your work aligned with field developments.

• Write clear documentation of methods, experiments, and findings to enable reproducibility and cross-team collaboration.

• Author technical papers and submit findings to premier conferences including NeurIPS, ICML, ICLR, CVPR, ACL, and AAAI to contribute to advancing model compression research.

What you'll bring

• A degree in Computer Science or related field, ideally a PhD in NLP, Machine Learning, or equivalent, supported by a proven publication record in top-tier conferences and hands-on AI research experience.

• Proficiency with PyTorch or comparable deep learning frameworks.

• Practical experience implementing both quantization-aware training and post-training quantization on real models.

• Demonstrated research and implementation background with knowledge distillation for model compression.

• Demonstrated research and implementation background with pruning techniques for model compression.

• Strong grasp of neural network fundamentals including transformer architectures, LLMs, VLMs, backpropagation, optimization algorithms, and fine-tuning workflows.

• Familiarity with C++ for low-level kernel development and inference optimization.