AI Engineer

Lead engineer owning our full-stack product and multi-agent AI end-to-end. Simplex 3D builds a SaaS platform for 3D urban and architectural planning, fusing photorealistic 3D, large geospatial (GIS) datasets, and a GenAI-native chat into one application for real-estate and urban-planning professionals. The Role A full-ownership role for an AI-native software developer who owns the full-stack product and designs, ships, and operates multi-agent systems end-to-end — both the outward-facing agents inside the product and the inward-facing agents that accelerate development — on a 3D geospatial platform. What You'll Own Full-stack development: own the React/TypeScript frontend and Java/Spring, Node.js and Python backend on AWS with PostgreSQL; extend the existing codebase independently with complex features across 3D, GIS layers, and CRM. Multi-agent chat (outward-facing): a natural-language assistant that operates the platform via MCP tools — layers, camera, map editing, model upload, presentation/report generation — with specialist agents reasoning over geospatial data (zoning, building rights, appraisals, transactions, permits) grounded in RAG and per-client memory. Dev-acceleration agents (inward-facing): AI-assisted CI/CD and dev-productivity agents (agentic coding, code review, test generation, build triage) plus internal knowledge and data-pipeline agents over our docs and geospatial schemas. MLOps / LLMOps: evaluation harnesses, tracing and observability, guardrails and prompt-injection defense, prompt/version and context management, and latency + cost optimization for production agents. 3D & GIS platform: advance 3D model conversion, geospatial data pipelines, tileset and layer loading, and AI-generated images/video for marketing. Requirements 5+ years of software engineering; BSc in Computer Science / Software Engineering or equivalent. Multi-agent systems: designing agent loops — tool/function calling, structured outputs, sub-agent orchestration, memory, and handoffs — with a production framework (LangGraph preferred; CrewAI, AutoGen, or OpenAI Agents SDK also relevant). MCP: building or consuming Model Context Protocol servers/tools, or designing rigid JSON tool-calling schemas for agents. LLM & RAG in production: Anthropic and/or OpenAI APIs; retrieval design, embeddings, reranking, and a vector database (e.g. Pinecone, Weaviate, Qdrant). AI-native workflow: daily use of agentic coding tools (e.g. Claude Code, Cursor, Copilot) and context engineering — managing memory, token budgets, and state. MLOps / LLMOps: evals and observability (e.g. LangSmith, Langfuse, Arize Phoenix), guardrails, prompt-injection defense, and latency + cost optimization. Cloud: AWS, Docker, CI/CD, and version control (GitHub). Independent, full-ownership engineer able to take features from problem to production alone. Full-stack: React and TypeScript/JavaScript; Python and Java/Spring Boot and/or Node.js; SQL/PostgreSQL. Advantages GIS / geospatial: map-based web apps and spatial data (PostGIS, Mapbox, GDAL). 3D graphics: WebGL, Three.js, or CesiumJS; 3D model conversion and tiling; gaming/engine frameworks (Unity, Unreal). Computer vision & 3D ML: reconstructing 3D from imagery and scans — Gaussian splatting (3DGS) and NeRF, models such as NVIDIA fVDB and Meta SAM 3D, plus photogrammetry, point clouds/LiDAR, semantic segmentation, and mesh reconstruction (PyTorch). Domain knowledge: hands-on experience with urban-planning or architecture projects and an understanding of planning, zoning, and design workflows. AI image/video generation for realistic renders and marketing content. A2A (Agent-to-Agent) protocol and additional frameworks (Google ADK, Semantic Kernel, LlamaIndex). Advanced evals (LLM-as-judge, RAGAS), sandboxed execution (E2B, Modal), and graph databases (Neo4j, Memgraph). Microservices, DevOps, and securing API interactions (auth, tokens, OAuth 2.1).