Apply on Employer Site

Xcede · 1 day ago

Machine Learning Engineer

San Francisco Bay Area

Full-time

Onsite

Mid, Senior Level

$200K/yr - $500K/yr

2+ years exp

Xcede is a company focused on advanced technologies, and they are seeking a Machine Learning Engineer to design, train, and deploy general embodied agents. The role involves building perception-to-action pipelines and developing scalable interaction infrastructure for large-scale training and evaluation.

AdvertisingHuman ResourcesInternetStaffing Agency

Hiring Manager

Isabel Zorn

Responsibilities

Design, train, and deploy general embodied agents integrating multimodal perception (vision, depth, language), reasoning (memory, planning, world models), and action (hierarchical control, continuous/discrete policies)

Build end-to-end perception-to-action pipelines, latent world models for UI environments, and predictive planning systems

Develop scalable interaction infrastructure, high-throughput data engines, and safe, sandboxed execution environments to support large-scale training, evaluation, and iteration

Action abstraction & control: Keyboard/mouse primitives, macro-actions, hierarchical action spaces. Latency-aware control loops; debouncing, key timing, cancellation. Continuous-to-discrete bridging (cursor trajectories → key events)

Perception → action pipelines: Screen-state modeling (pixels, OCR, DOM/UI trees, accessibility APIs). Temporal state aggregation (frame stacking, event deltas, change detection). Robustness to resolution changes, lag, UI noise, animations

World modeling for UI environments: Latent dynamics over screen states and UI affordances. Predictive models for action consequences (what changes if I press X?). Counterfactual rollouts for planning and error recovery

High-throughput interaction pipelines: Parallel rollouts with synchronized input/output capture. Deterministic replay for debugging agent behavior. Rich telemetry: action traces, screen diffs, attention maps

Isolation & safety: Sandboxed OS / VM environments Input gating and rate limiting Strict separation between agent policy and host system privileges

Data engines: Trace mining from human and agent play. Automatic dataset generation from interaction logs. Versioned datasets tied to behavioral deltas

Qualification

Embodied agent systems designMultimodal reasoningComputer VisionScalable product developmentData enginesTechnical fundamentalsAcademic achievements

Required

2 - 10 years of experience from competitive post graduate programs or high growth startups

Demonstrated track record building scalable products

Experience with agents and multimodal reasoning or diffusion models

Knowledge of Computer Vision or graphics

Signs of academic achievements

Designed and trained general embodied agents that perceive (vision, depth, language), reason (memory, planning), and act (continuous/discrete control)

Strong technical fundamentals across embodied agents systems design and creating systems and infrastructure from scratch