← All roles

Full-time · New York City, in office

Member of Technical Staff (AI Researcher / Engineer)

You will own every model Plurall AI ships: GaussMass for images, Influx for video, Sage for live calls and Timbre for audio, and the fusion that turns their evidence into one calibrated verdict. Research, training, evaluation and release are yours, from the first equation to the number a fraud analyst acts on.

Total compensation$1,000,000

What you will own

  • The four detection models and their roadmap: architecture, training data, release criteria and retirement
  • The evidence functions behind each verdict: spectral and noise-residual analysis for images, temporal and audio-visual consistency for video, liveness and injection checks for live sessions, vocoder artefacts for speech
  • Evidence fusion and calibration, so that a reported likelihood means what it says on real traffic
  • The measurement programme: named slices, confidence intervals, cross-dataset checks, and results reported by slice and never as one pooled accuracy
  • The retraining loop that takes a new generator family from first sighting to a re-scored release in 24 to 48 hours

What you will do

  • Design, train and ship models that hold up against generators that did not exist when they were trained
  • Prove what you build: write down why an evidence function should work before you measure whether it does
  • Keep models small and fast enough for real time, including on-device and NPU inference
  • Hunt your own failures: benign edits, re-encoding, screenshots, compression, adversarial post-processing and distribution shift
  • Decide when the system should say it does not know, and send those cases to review
  • Publish what can be published, including the next edition of our technical white paper
  • Work with forward deployed engineers to turn failures found in the field into new slices and better models

What you bring

  • A record of taking models from idea to production in computer vision, speech or multimodal learning, and of owning them afterwards
  • Depth in at least one of: image and video forensics, signal processing and harmonic analysis, generative models (diffusion, GAN, flow matching), speech synthesis and anti-spoofing, or face and liveness biometrics
  • Fluency in the mathematics the work rests on: probability and decision theory, calibration, sequential testing, and spectral methods
  • Rigor about evaluation: leakage, disjoint samples, shifted distributions, and what a number does and does not show
  • The engineering to match: you write the training pipeline, the inference path and the tests yourself
  • Publications, open-source work or shipped systems that we can read

Where you will work, and what we offer

  • In our New York City office
  • You live in New York City already, or you are willing to relocate
  • Total compensation of $1,000,000
  • Company benefits

Application