ML Engineer
Eagle
Posted about 2 hours ago
About Eagle
We’re on a mission to radically transform the way we design and construct our built environment.
Backed by Lightspeed Venture Partners, Eagle acquires and transforms civil, structural, and MEP engineering firms with applied AI. We’re an AI laboratory dedicated to providing engineers with the tools they need to solve the world’s hardest infrastructure, energy, and climate problems.
By arming designers with frontier technology, our ambition is to build the most valuable, talent-dense engineering firm in the United States.
The opportunity
Our core thesis: 85% of what engineers do today is theoretically automatable, yet less than 5% has actually been touched by AI. That gap is the largest of any profession. Our plan is to close it by acquiring engineering firms, building purpose-built tools for their staff, and compounding that proprietary intelligence across acquisitions.
The richest, most defensible data in this industry lives in 2D drawings—drawings sets, details, sections, schedules—and only a small fraction of it is machine-readable today. As a Machine Learning Engineer, you'll own the problem of turning that visual information into structured, embedded, queryable intelligence. You'll work directly with the CTO, and the work you do becomes the foundation the rest of the platform compounds on top of. You get a front-row seat to building a company from zero—engaging with architecture decisions, firm acquisitions, and product strategy—on a problem domain that's barely been touched by AI.
What you'll do
Embed with staff at engineering firms alongside the founders; get your hands on real drawing sets and learn how engineers actually read, mark up, and reuse them
Own the drawing-parsing pipeline end-to-end—ingestion of PDF and CAD exports, layout analysis, symbol and entity detection, OCR on dimensions and notes, and extraction of schedules and title-block metadata from noisy, inconsistent real-world sheets
Design the embedding strategy for drawings: how to represent a sheet, a detail, or a region as a vector so it can be searched, compared, and reasoned over—adapting or fine-tuning vision and multimodal encoders as needed
Integrate extracted structure and embeddings into our knowledge store so it gets richer and more valuable with every drawing and every acquisition
Build the evaluation harness this all depends on—ground-truth sets, accuracy metrics, and a tight loop for measuring whether the models actually work on messy production data
Collaborate directly with the CTO on technical direction and what we'll build next
What we look for
Deep computer vision and VLM experience, ideally on documents, diagrams, or drawings rather than only natural images—detection, segmentation, layout analysis, OCR
Wants to obsess over this high-leverage data problem: pulling signal out of drawings that were never designed to be parsed by a machine
Understands embeddings and representation learning—how to build, fine-tune, and evaluate an embedding space, not just call an API
Ships to production and owns the result; this is an engineering role, not a research-only one
Has the rigor to be honest about model quality on real data, and to build the evals that keep everyone honest
Has a deep curiosity for how things work (an organization, a workflow, a market)
Isn't afraid to expose their ignorance and is constantly asking why
Has the poise and communication skills to earn trust with people who've never worked with a tech company before
Is willing to get on a plane with us
Is not above any task: up to label the data yourself, write the annotation tooling, or hand-tune a heuristic when the model isn't ready yet
Compensation
Competitive cash compensation ($150K–$300K depending on experience)
Meaningful early-stage equity
Full healthcare benefits
In-person office in NYC
