Arvist (Techstars 2022) is hiring ML/Back end Developers and Consultants

The Backend/ML Engineer role brings Arvist’s customer value together by combining streams from RTSP protocols with ML to create reportable data using real-time feed via ML models and mathematical analysis (geometry, algebra, and classic programming). ML models are only one tool used to detect anomalies, and must be used in conjunction with mathematics and potentially consecutive ML models to define thresholds, distances, and events (observations, reports, etc.) within Arvist’s infrastructure. These processes help partners customize and define processes within a worker facility to create a digital twin system. Customers define which regions of an IP camera are processed by specific ML models and workflows to reduce redundancy and mismarked observations. Customers can define interval and schedule based tasks in a generalizable way (via celery beat) to control how often a workflow is run on specific cameras (fps) and if it is only run during certain operating hours (for fine-grained control of operations). A primary goal for Arvist is for any workflow we create for customers to be generalizable to any camera with independent scheduling and regions within its related camera(s).

Scope (Backend): Python HTTP API’s (FastAPI) Python Task Management (Celery + Celery Beat) Asynchronous Programming Postgres (asyncpg) Processing RTSP frames (open-cv)

Scope (ML - Image Segmentation, Image Classification, and Object Detection): Hugging Face Transformers Pipeline (for hybrid on-prem/cloud models) AWS Sagemaker Inference or Google AI Platform (for cloud-only models) Developing a range of computer vision based models with fast turnaround time Processing and refining datasets for creating generalizable and accurate ml models Experience utilizing geometry, classical programming, and vision-based mathematics to produce reports and observations for customers that hold value or non-obvious insight. This means utilizing a deployed model within our backend service (FastAPI, Celery, and Celery Beat) to bridge model inference with insightful behavior.

Other Tools: Local Dev Tools Docker (and Docker Compose) Git (Github) GNU Make Redis (will be nice for cacheable results like seeing how objects moved without contacting a database)

Examples and Upcoming Work: How long was a worker inside a room? Where did the pallet at 11:00UTC get dropped off at? What room did it end? Did it ship successfully? Was the digitized process aligned with what the managers intended? Was a semi parked longer than initially estimated? Did someone forget about it? Detecting spills in rooms, what time was it cleaned up? Between the hours of 19 and 6, was anyone detected not wearing an authorized jacket (yellow vest) Were any pallets left outside their designated stacking zone?