SDE 2/3 AI Engineer, Digital Production (GenAI 3D Agents)

Role Overview

Metadome.ai builds frontier AI that turns text, drawings, and CAD into production-ready, interactive 3D — powering immersive product experiences and aftersales modernization for 30+ automotive and heavy-equipment manufacturers.

We're now building a new marketing AI platform that replaces days of manual 3D content work with generative models and agents that drive tools like Blender and Unreal Engine directly. As one of the founding engineers on this platform, you'll set the technical bar for how it's built — agentic workflows, AI-assisted development, aggressive automation, tight eval loops — with a direct line to the CTO.

This role calls for a 100x, agentic mindset: if you've done something twice by hand, you script it or build an agent for it. You treat AI tools as force multipliers and own their output completely. Prototype to production is one motion for you — ship fast, measure, iterate — and you default to ownership, finding problems and closing them without being asked.

Responsibilities

• Own generative AI, computer-vision, and diffusion-model features end-to-end, carrying each one from prototype through production, evaluation, and iteration.
• Design and ship agentic systems — built on LangGraph and MCP — that drive Blender and Unreal Engine through Python to automate asset preparation, material assignment, shot composition, lighting, and rendering.
• Build the infrastructure around the models: model serving and backend services and APIs in Python (FastAPI), Go, and Node.js, deployed on GCP (Cloud Run, serverless, and GPU workers) and backed by PostgreSQL, MongoDB, Redis, and Neo4j — with evals powered by RAG pipelines and LangSmith/Ragas, observability through Prometheus and Grafana, and modern web frontends where needed.
• Work directly with 3D and geometric data — meshes, materials, cameras, lighting, and render pipelines — to bring the pipeline to production quality.
• Break ambiguous, research-flavored problems into shippable increments, measuring quality with evals rather than intuition.
• Raise the technical bar around you as a hands-on individual contributor — through code reviews, internal demos, mentoring, and disciplined CI/CD practice via GitHub Actions and Cloud Build.

Requirements

Experience & Background

• 3–7 years shipping and maintaining production software, with a track record of end-to-end ownership.
• Hands-on production experience with LLM systems (RAG, tool use / function calling, agents) and/or diffusion or vision models — shipped systems, not notebooks.
• Strong grounding in training, evaluating, and deploying models against real quality and latency bars.
• Proficiency in Python (FastAPI) plus Go or Node.js, a modern frontend framework, cloud experience (GCP preferred), and solid CI/CD habits.
• Daily use of AI coding agents (Claude Code, Cursor, Copilot) — with rigorous verification, testing, and ownership of every line you ship.
• Comfort building golden datasets, grading trajectory vs. outcome, and enforcing quality gates in CI.

Tools & Technologies

Core Stack

• Python (FastAPI), Go, Node.js
• GCP — Cloud Run, serverless, GPU workers
• LangGraph, MCP, RAG + evals (LangSmith/Ragas)
• Diffusion & vision models
• Blender & Unreal Engine
• PostgreSQL, MongoDB, Redis, Neo4j
• Prometheus/Grafana, GitHub Actions/Cloud Build

Good to Have

• 3D/graphics programming: WebGL/OpenGL, GLSL shaders, rendering pipelines, performance profiling (draw calls, instancing, culling)
• Blender or Unreal Engine Python scripting; any DCC/3D pipeline automation
• Generative 3D/CAD exposure: text-to-CAD, 2D-to-3D reconstruction, controllable/conditioned generation
• Backend depth: resiliency patterns (circuit breakers, retries), security-first API design (OAuth2/JWT), GitOps
• Research output, patents, hackathon wins, or open-source contributions

Role Benefits & Growth

You will take founding-level ownership of a brand-new AI product with real enterprise pull — 30+ OEM customers from day one.

This is an AI-native environment by default, with model access and compute budgets built in and full backing to reach for the best available tools — a clean slate with no legacy system to maintain, and the opportunity to help define the engineering patterns everyone who joins after you will follow.