Research Engineer - Decentralized Training and Inference Verification
Pluralis Research · Remote · San Francisco · Australia ·
COMPENSATION
€•• – ••k
Listed on the employer's job board (Ashby)
- Fully remote (global)
- Visa sponsorship available
- Senior research role
- Statistical verification focus
- Decentralized AI security
Pluralis Research is advancing Protocol Learning, enabling large model training and serving on decentralized consumer devices. This role focuses on designing efficient algorithms to verify that inference tokens and training contributions are legitimate in a trustless environment.
You will define threat models, build statistical verification methods with controlled error rates, and deploy these services across heterogeneous hardware. The team operates globally, primarily from Australia and North America, tackling open problems in decentralized ML security.
What you'll do
- Define and maintain threat models for malicious or careless workers in pre-training, post-training, and inference.
- Design statistical verification methods with stated error rates and tune them via rigorous benchmarks.
- Control false positives and negatives across diverse hardware including various GPUs and Macs.
- Build and operate the verification service within the inference path.
- Address attacks such as free-riding, model poisoning, data extraction, and reward manipulation.
What you'll need
- Proven experience building and shipping calibrated statistical decision systems with known error rates.
- Deep expertise in statistics and probability, including experiment design and threshold calibration.
- Knowledge of solutions for verifying untrusted compute, such as cryptographic proofs or trusted hardware.
- Ability to argue which verification techniques fit a permissionless network context.
- Professional-level proficiency in written and spoken English.
Nice to have
- Familiarity with large-scale pre-training and RL post-training processes.
- Experience with decentralized ML security threats like Sybil attacks, collusion, and poisoning.
- Background in proprietary, open-weight, or open-source AI laboratories.
- Publications in inference and training verification or related fraud detection fields.
Stack
- Statistical testing
- Cryptographic proofs
- Trusted hardware
- Re-execution methods
How you'll work
- Fully remote position with team members distributed globally.
- Main teams located in Australia and North America.
- Must be comfortable working across multiple time zones.
- Flexible work environment.
What you get
- Equity-heavy compensation package with significant ownership opportunities.
- High base salary.
- Optional full visa sponsorship and relocation support to Australia or the US.
- Opportunity to solve unpublished problems in decentralized AI infrastructure.