NVIDIA Omniverse / OpenUSD Engineer — AI-Prompt Scene Control (Start This Week)

We are a multinational studio with centers across Latin America and Pakistan, working on real-time 3D and visualization projects. Our lead technology team is distributed across Latin America and Pakistan. This is a rapid prototype for a client with a tight, fixed budget who is genuinely interested in the technique — using AI prompts to drive a live NVIDIA Omniverse scene and generate camera fly-throughs. We're being upfront that the budget is lean; what we're offering is a small, well-defined, interesting problem with a clean scope and a fast finish. We already have the Omniverse environment set up — we need a specialist to finish the last mile. Delivery is needed by this Sunday at the latest, so you must have hours free this week. What we already have (your starting point) A cloud virtual machine with the required RTX GPU, accessible remotely (RDP/Parsec). NVIDIA Omniverse installed and running (USD Composer / Kit-based app). An OpenUSD (.usd) model of a live-event stage, already loaded. Our own AI/agent stack (we drive tools with Claude) and a technology lead + a creative partner who will pair with you throughout. What we need you to build (be prepared to do exactly this) 1. An AI-to-Omniverse control bridge. Build a small Omniverse Kit extension (Python) that runs inside the live app, opens a local socket, and executes USD operations on the running stage on command — following the well-known socket + execute_python pattern (reference: the open-source omni-mcp/isaac-sim-mcp). Pair it with a thin MCP server so our AI agent (Claude, via mcp.json) can call it. Handle the main-thread execution correctly (USD edits marshalled onto Kit's update loop). Include log read-back after each call so the agent can self-correct. 2. Make the scene addressable. Expose helpers to list prims (path, type, bounding box, world position), capture a viewport screenshot, and set displayName / semantic labels so elements can be targeted by name ("main screen," "key light," "rail cam"). We will run an agent pass to name the model; you provide the tools that make it possible. 3. Prove core live commands (record them working): Move a named object a set distance ("move the main screen 10 ft stage-left"). Adjust lighting from a reference (set light color/intensity/position). Switch the active camera/view. 4. A prompt-driven camera fly-through, rendered at 1080p. Set up camera paths (Motion Path / Sequencer) and Movie Capture output so we can generate fly-through shots and render them to 1080p video/image sequences. Target a handful of short shots (a few seconds each) we can cut into a sequence. Benchmark one full-quality frame on our VM early and give us a real per-frame render time so we can plan. (We are targeting 1080p, not 4K, for this round.) 5. Handover. Leave it runnable by our team — a short written + Loom walkthrough of how to issue prompts and kick off a render. Estimated effort for someone fluent in Omniverse Kit: ~20–28 hours. If it's taking dramatically longer, this may not be the right fit. How you'll work You will pair with one of our technology leads and a member of our creative team. The creative owns shot design and the lighting look — you do not need a strong cinematography background. Fly-through / virtual-camera experience is a plus, but your core job is the engineering: the bridge, the scene tooling, and the render pipeline. What we are NOT looking for Someone who only does offline/traditional rendering and can't script inside Omniverse Kit. A pure 3D modeler with no Python / real-time-engine experience. Anyone who needs weeks to ramp — this is a days, deliver-by-Sunday engagement. Requirements NVIDIA Omniverse / Kit — hands-on. Kit extension experience strongly preferred (tell us what you've shipped). OpenUSD — comfortable with prims, transforms, layers, cameras, Movie Capture. Python — the bridge and all scripting are Python. Real-time 3D — Unreal/Unity/technical-artist background counts. Comfort with AI coding agents / MCP (Claude, Cursor) — you'll pair with an agent and our lead. English fluency required. Spanish preferred. Candidates based in Pakistan especially welcome; international candidates welcome. Availability this week, delivery by Sunday — please don't apply unless you can commit the hours now. To Apply Start your cover letter with the phrase "USD Bridge" and tell us: (1) whether you've built or extended anything in Omniverse Kit (or a comparable real-time engine) and what it did, and (2) confirm you have hours this week to deliver by Sunday. One short paragraph — we value signal over length.