Thomas Hossler

Part-time contracting and advising: * Zanskar Geothermal: infrastructure work (kubernetes, gcp, terraform), R&D (geospatial AI), project lead * Lumetec: system design, cloud computing

Building Vitruvius, the first AI architect https://iconbuild.com/vitruvius (project cancelled and laid off). Using LLM and diffusion models to generate blueprints.

* MLOps: skaffold, terraform, kustomize, helm, kubernetes * MLinfra: triton, polygraphy * R&D: object detection, sensor fusion

Designed and built content for the Self-Driving Car Nanodegree. Course on Deep Learning for Computer Vision

I joined Aquabyte as employee #3 as the main computer vision engineer. I helped grow the Aquabyte team by a factor of 5. Aquabyte is revolutionizing the aquaculture industry by improving precision fish farming through computer vision & deep learning algorithms. Using our dashboard, farmers have a better understanding of the current status of the fish population (average biomass, lice infestation) and can make relevant business decisions. - As the first deep learning engineer, I built most of the computer vision algorithms - object detection, stereo matching, image classification, keypoints detection (Tensorflow, Keras, Pytorch) as well as some machine learning algorithms such as biomass estimation (sklearn) - I worked on the rectification and calibration pipelines of our stereo camera system (OpenCV) - I also designed machine learning pipelines, from field experiments to the cloud infrastructure (AWS: SQS, SNS, S3, EBS). I worked on supervised and unsupervised algorithms, using observed and synthetic data (Blender, Labelbox) These algorithms are being used in every products at Aquabyte. - I interviewed over 50 engineers for various positions (backend, edge, ML).

Focal Systems is a Stanford-born deep learning company helping retailers tackle their biggest challenges through deep learning and computer vision. I joined Focal Systems pre-Series A as a Software Engineering Intern, working on algorithms and data visualization. I then joined the Deep Learning team and I lead the project on the indoor positioning system using sensor fusion. I also researched an algorithm for general purpose template matching. - worked on automated checkout system using synthetic data (training using Keras, deployment on Jetson TX1) - took part of the development of an indoor localization system using image classification and wifi signal strength (Keras, hidden markov models). - build a data visualization system to monitor the heartbeats of our field devices (Superset, bokeh, MongoDB, SQL) - helped maintaining cloud infrastructure (GCP, docker)

Second year PhD student in Jef Caer's group, the Stanford Center for Reservoir Forecasting. The group interests lie in making decisions and forecasting for reservoirs and groundwater. Dealing with high dimensional uncertain environments, our research uses geostatistics and machine learning technics.

Multi-point statistics, geostatistics, Image Quilting algorithm

Fieldwork in Eastern Nepal (Surkhet), paleosismology, geomorphology, tectonics

Eddy tracking algorithm, Sea Surface Temperature/Height data sets, Principal Component Analysis