AI Engineer (Computer Vision) - Full Time - NA/LATAM

The Role

We are looking for an AI Engineer (Computer Vision) to join an existing codebase and ship features end-to-end for an AI-assisted roof estimation MVP that ingests PDF blueprints, extracts roof geometry (deterministic + VLM), provides a raw SVG HITL board for reviewer corrections, and generates verified takeoff reports.

Tech Stack

• Language: JavaScript (ES2020+); Node.js runtime (Node 22, pinned 22.22.1 via .nvmrc)

• Backend: express for HTTP server; server entry at index.js

• PDF processing: pdfjs-dist (rasterization + PDF parsing); helper scripts in pdfDocuments.js and pdfPreRasterService.js

• Deterministic extractor / Roof graph: custom extraction code in roofGraph (extractor, constants, worker)

• Frontend: Vanilla JavaScript (no bundler); orchestrator in main.js

• HITL board: SVG-based interactive board

• LLM / VLM Providers: OpenAI + Google Gemini adapters under providers and a proxy route at llm.js

• Validation: zod used across server & client contracts (chat contracts in contracts)

• Database / Storage: better-sqlite3 for local persistence (db.js); storage abstraction in storage.js with optional @aws-sdk/client-s3 for R2/S3

• Auth/Session: JWT-based session management (auth.js) with httpOnly cookie rp_session

• Testing & QA hooks: Browser QA hooks (browserQaHooks.js) expose window.__roofProofTestHooks for headless testing without altering product behavior

• Dev / CI / Containers: DevContainer / Docker Compose (docker-compose.yml) for local development; npm scripts for dev, checks, and verifiers; production deploys to AWS via GitHub Actions (OIDC → ECR → EBS redeploy via SSM); persistent EBS volume for SQLite and uploaded PDFs

• Other tools: Node --watch dev flow, deterministic milestone verifiers, some Python used for experimental PDF prepass tooling (pdf_prepass)

• Notes: Codebase mixes deterministic image-processing with VLM-assisted flows; many flows follow a contract+flow pattern (flows + contracts).

Major Responsibilities

• Maintain and extend the deterministic roof extraction algorithms in roofGraph

• Build portal foundation for proper authentication and access controls.

• Implement and validate VLM-assisted flows (contract + flow) and their server adapters (providers, llm.js, chat)

• Improve the Fabric.js board UX and synchronization with backend state (board, boardStore.js)

• Add/maintain verification scripts and QA hooks used by npm run verify:*

• Hardening: input validation (Zod), error handling, and performance/scalability for rasterization and extraction pipelines

• Work with Docker/DevContainer for reproducible development and CI integration

• Write tests or verifiers for new contract surfaces and update verify-baseline.js when adding milestone verifiers

• Collaborate on deployment concerns (optional): S3/R2 storage, ECR/terraform infra in aws

• Suggest enhancements and restructuring for better performance from the agentic workflow

Required Experience & Skills

• Experience with full-stack JavaScript/Node.js development (Express)

• Strong experience with canvas libraries (Fabric.js or equivalent)

• Familiarity with PDF rasterization and vector/image processing (pdfjs-dist or similar)

• Experience integrating LLM/VLM providers and designing robust request/response contracts

• Practical SQLite experience (better-sqlite3) and schema/transaction design

• Proficiency with Zod or runtime schema validation patterns

• Comfortable with Docker / DevContainer and npm-based workflows

• Good debugging skills for long-running worker processes and child-process architectures

• Test automation experience (Playwright or similar) is a plus

Nice-to-Have

• Prior experience with OpenAI/Gemini and prompt/response engineering for multimodal flows

• Experience with AWS S3 or Cloudflare R2 and @aws-sdk/client-s3

• Background in geometry, computational geometry, or CAD-like tooling

• Familiarity with TypeScript migration (repo currently uses vanilla JS)

• Experience writing verification/milestone scripts and CI workflows