Member of Technical Staff: Junior AI Researcher

Synhawk•San Francisco, CA

19h

About The Position

Synhawk builds omnimodal foundation models for communication integrity, aimed at infrastructure-side deployment in telco and banking sectors. Our platform analyzes the integrity of audio and video, and protects platforms against AI threats. This currently includes detecting synthetic speech & voice cloning, video & image manipulation, social engineering, and identity impersonation. We’re building our models in a way that generalizes to future AI threats yet to come, especially as AI agents get integrated into the workforce and society. We're past the initial research phase. We're actively deploying with major telco customers, working with and building our own GPU clusters, air-gapped environments, and strict production SLAs. Your work with us will have an immediate, measurable impact on systems that defend real communications infrastructure. We're a small, highly technical, founder-led team. You'll play an integral part in shaping our research agenda while building foundation models to help real customers from day one. The Role You'll own and advance Synhawk's core engine - spanning the development of foundation model and media integrity methods in a full research-to-production pipeline. This isn't a pure research role, as the models and methods you build will directly feed into production with our telco partners. That is to say, you’ll be working under real latency and reliability constraints. But no pressure. While practical, real-world applicability grounds how we build, we also invest deeply in forward-looking research into emerging AI communication threats. In practice, we balance both: staying focused on protecting customers today while preparing for threats that don’t yet exist. You'll have direct influence over our research agenda from day one. As we scale beyond early deployments, this role also provides opportunities to publish novel research and engage with the broader community through conferences and events.

Requirements

Finishing a Master's or PhD, or recently graduated, with strong foundations in machine learning and a real interest in audio, video, or media forensics.
Comfortable in Python and PyTorch
Experience training models yourself
Experience with data, debugging, and analyzing experiment results

Nice To Haves

A thesis, a paper, open-source contributions, a research internship, or side projects
Eagerness to become an expert in the required areas

Responsibilities

Running and scaling training experiments on multi-GPU setups
Data curation: filtering, deduplication, and building training data mixtures
Fine-tuning and post-training (SFT, preference optimization), and evaluating how models behave under distribution shift
Multimodal data augmentation strategies
Implementing and testing detection architectures for audio, video, and images
Signal processing: codecs, compression artifacts, forensic fingerprints
Adversarial robustness: testing models against attacks, re-compression, and downsampling
Hands-on red teaming with open-source and proprietary deepfake generators, integrated into the R&D cycle