Computer Vision Research Engineer

San Franciscofull-timemid

OtherResearch Engineer

0 views0 saves0 applied

Apply Now

Quick Summary

Overview

Position Overview Bobyard solves non-trivial CV problems that move millions of dollars. Contractors burn dozens of hours per project doing takeoffs by hand. We automate it — with models that have to be better than human, on messy real-world drawings.

Key Responsibilities

Build SOTA CV models: Detect, segment, OCR, and reason over construction drawings. Accuracy isn’t academic — one mistake costs a customer millions. Own end-to-end pipelines: Data ingestion, labeling, training, eval, and low-latency inference.

Technical Tools

segment

Bobyard solves non-trivial CV problems that move millions of dollars. Contractors burn dozens of hours per project doing takeoffs by hand. We automate it — with models that have to be better than human, on messy real-world drawings.

You’ll have founder-level autonomy to design, execute, and iterate. We move at startup speed — no research silos, no “publish and forget.” You’ll become the world’s leading expert on the models you ship. We want world-class ML engineers with product taste and the ability to go heads down and produce exceptional systems.

Responsibilities

~1 min read

→
Build SOTA CV models: Detect, segment, OCR, and reason over construction drawings. Accuracy isn’t academic — one mistake costs a customer millions.
→
Own end-to-end pipelines: Data ingestion, labeling, training, eval, and low-latency inference. If it doesn’t run in prod, it doesn’t count.
→
Scale performance: Accuracy, robustness, latency. Our models face 10,000+ drawing formats and they need to just work.
→
Ship to production: Work closely with fullstack and product to integrate models into features customers use daily. No throwaway notebooks.
→
Attack failure modes: Dig into real customer data. Find where we break, why we break, and fix it fast.
→
Apply research pragmatically: Stay current with the field. Take what works, ship it next week, measure impact.

Founder-level drive: You own model performance like it’s your company. No one tells you what to fix — you already know.
Deep CV + DL chops: You’ve trained models that actually shipped. Segmentation, detection, document AI, vision-language — you’ve been in the trenches.
Production killer: You’ve put ML into real products with real SLAs. You know the gap between 90% and 99.9% and how to close it.
Messy data mindset: Real drawings are crooked, scanned, handwritten, and ugly. You love that. You turn noise into signal.
Research-minded, execution-focused: You read papers, but you ship code. You bias to impact over elegance.
Startup work ethic: This isn’t a research lab or a cushy big-tech role. We’re at war with manual takeoffs in a $13T industry. You want that.
Learning machine: New architecture on Monday, beating our baseline by Friday. You navigate unknown problem spaces fast.

This is full-time & in-person in the SF Bay Area. Learning rate and dedication beat pedigree. If you can prove you’ll ship models our customers trust with million-dollar bids — at the speed and quality the market demands — we want you.