Founding AI Engineer (Multimodal Ingestion & Agents)
What We’re Building
Candid is collapsing pre-construction engineering for large compute and energy infrastructure projects.
Today, critical work lives inside complex diagrams, specs, and technical documents and is still done manually. We’re building systems that can understand this data, structure it correctly, and execute workflows like design review, comment resolution, and technical bid evaluation.
What You’ll Do
Build vision pipelines that extract structure, entities, and spatial relationships from engineering diagrams and documents.
Design agentic systems that decompose large pre-construction tasks into discrete, verifiable steps.
Build human-in-the-loop flows to capture expert feedback and improve system correctness.
Own systems end-to-end, from raw inputs to structured outputs used in production workflows.
Tech Stack
Python, PyTorch, multimodal LLMs, and open-source VLMs (Gemma, Qwen, DeepSeek, etc)
LangGraph, PydanticAI, or custom state machines
Graph databases (Neo4j), vector databases
Desirable Quirks
Strong opinions about system design, even when tradeoffs aren’t obvious
Cares deeply about keeping the codebase clean and understandable
Feels physical discomfort when seeing inefficient code or fragile architectures
Has good intuition when designing new systems without a spec
Cares about taste, clarity, and building things that feel right, not just things that work
Desirable Experience
(any of these will pique our interest)
Early or founding engineer at a YC-backed company
Experience with document intelligence, vision pipelines, or agent-based systems
Built something from scratch that real users depended on
Technical founder at a VC-backed startup