Engagement: project-based, fully remote · contract from 3-6 months with possible extension for future work · direct monthly payment (no HR / office / provided hardware) · English required.
About the role
ATMOS builds computer-vision safety systems for construction sites: dozens of on-site cameras, automatic real-time detection of unsafe situations, and easy human verification. We’re looking for a self-driven Full-Stack & AI Engineer to own the product end-to-end — from the camera video stream to a production-ready solution.
This is a one-person process: you work directly with a manager on our side, with no team-based development. You own the architecture, the implementation, and the quality of the result. Most of the work is backend and the AI pipeline, but you must also be able to build the frontend (expected to be done with the help of AI tools).
What you’ll do
Turn live camera video into frames and run batch video processing — sequential, stable, no crashes.
Computer Vision with pre-trained models (e.g. YOLOv8): object detection and unique-person tracking with persistent IDs (ByteTrack / SORT / model.track) — not frame-by-frame detection, but counting unique individuals over time.
AI-based safety-event analysis with reasoning: not just “a person is present,” but “a person under a crane load — risk.”
Real-time dashboard: live processing-status updates (WebSockets / SSE / efficient polling), aggregate metrics, a gallery of representative frames.
Camera integration via SDK/API: pull a frame or a video segment by timestamp (VOD for verification), automatic PTZ presets (sweeping site zones).
Human-verification UI: a clean interface to confirm “event / not an event.”
Design for scale — up to dozens of concurrent RTSP streams in production.
Must-have
A very strong engineer, willing to dig into both backend and frontend (the bulk of the work is backend).
Substantial server-side experience and cloud, especially AWS (real hands-on service work, not “heard of it”).
Computer Vision / Vision AI — using pre-trained models for detection/tracking (not training from scratch); solid grasp of object tracking.
Confident Python (the typical stack for CV/ML pipelines) plus a modern web stack for the dashboard.
Real-time architectures (WebSockets/SSE), video/media processing, queues/workers.
A mature AI-first approach: daily work with Cursor / Claude Code and similar — not just “I use it,” but the ability to drive agents well and take full ownership of the generated code.
Ability to deliver solutions end-to-end — from architecture to a working product — independently.
English — confident, for daily written and spoken communication.
Nice to have
Experience with RTSP / PTZ cameras, video streaming, media pipelines (ffmpeg, MinIO/S3, codecs).
MLOps / ML-inference deployment, GPU inference, latency optimization.
Geospatial / working with zone coordinates, edge processing.
Product work at startup pace and fast integration of solutions.
What we offer
A meaningful engineering challenge at the intersection of CV, AI, and real-time systems, with real impact (worker safety on construction sites).
Full remote and flexible hours — outcomes matter, not clocked time.
Direct collaboration with minimal bureaucracy and room to make technical decisions yourself.
A contract with the potential to extend into future work.
Take-home for this role: in 5–7 hours, build a working video-processing web app. In other words, we’re looking for exactly the person who can confidently deliver this — and grow it into a full product.
