Andrew Root

San Francisco, CA

• Built and shipped an end-to-end vision/audio system spanning sensors, edge devices running real-time inference, and AWS cloud infrastructure, then deployed it into dental operatory rooms across 2 customer offices; currently handling 10-20 appointments a day.
• Built an agentic inference pipeline on AWS that turns operatory audio and video into procedure predictions for insurance claims, combining speech-to-text with LLM calls constrained to structured output so that every step stays machine-parseable.
• Led the installs at the office I support today and collaborated with the clinical staff who use the system during live treatment; currently operate 10 edge devices across 5 operatories, each room pairing an Orange Pi 5B with a Raspberry Pi Pico 2W.
• Cut failure detection from a customer phone call to an automatic alert by building the deployment's observability layer on AWS CloudWatch and SNS: structured, queryable logs, and alerting on device and pipeline faults.
• Turned installation from undocumented practice into a repeatable checklist the team follows, authoring both documents Daentra installs from: a deployment guide covering supply lists and in-room installation, and a provisioning runbook for flashing, AWS device registration, and validation.
• Redesigned the in-room hardware install after the first office reported cabling interfering with treatment, rerouting wiring through the dental light and repowering the Orange Pi, camera, and Pico from existing operatory circuits within the supply's ratings.
• Designed a frame-timing synchronization algorithm that aligns a machine vision camera to a high-intensity dental light flickering at varying frequency and color.
• Optimized the edge devices, working closely with hardware, to stream 1200p video at 55 FPS on a single CPU core, offloading RetinaFace to the NPU and video encode/decode to the VPU; wrote a custom UVC kernel module to raise buffer capacity.
• Researched and prototyped depth and pose estimation methods for key-object tracking in video, benchmarking state-of-the-art models against custom implementations.

Rochester Hills, MI

• Shipped a real-time computer vision pipeline (Python, PyTorch, OpenCV) that segments expected objects from live video for an automated security vehicle system, in collaboration with Dataspeed Inc.
• Collected and curated the training data, and collaborated with Master's and PhD students on evaluation.
• Built deep learning, PyTorch, and computer vision expertise independently, starting from no prior background in the field.

Eau Claire, WI

• Quantified how far cyberbullying-detection models degrade on data they were never trained on, running cross-dataset evaluations to isolate sources of bias in ML datasets; mentored junior researchers.
• Ran long-running training and evaluation workloads with XGBoost and decision trees on UWEC's high-performance computing clusters.

Eau Claire, WI

• Put advanced SQL database operations in the hands of non-technical manufacturing teams by building and deploying a full-stack web application.
• Refactored 3 production manufacturing services integrating PLC control systems and message brokers.
• Prototyped an auto-documentation service that generates Confluence pages from code repositories using web scraping and syntax parsers.

An agent-first relationship layer over Git, published to PyPI as tether-it. A tether links two artifacts, such as a function and the documentation describing it, fingerprinting both ends with git blob OIDs so that drift is flagged at the symbol or section level rather than across the whole file. Paths that break are run through git's rename detector to surface a rename candidate. Ships a Claude Code integration: session hooks that report drift, permission rules, and an onboarding skill that seeds the initial graph.

Python · Git · PyPI · GitHub Actions · Claude Code

A full-stack, real-time voice agent that lets users create, edit, and manage tasks entirely by speech, with no traditional UI controls. Audio streams to a Python agent that transcribes it, resolves intent with an LLM, executes the matching database operations, and pushes updates to the frontend live.

Python · LLMs · LiveKit · Speech-to-Text · Next.js · TypeScript

Led the university's ACM chapter: designed and ran hands-on sessions on Git/GitHub, Python, and neural networks, chaired meetings, and delegated tasks to a board of officers.

Designed and taught a workshop to 20+ UWEC students covering Python, statistics, data analytics, and machine learning.

B.S. Computer Science & Applied Mathematics (Double Major)

• GPA: 3.98 / 4.0
• Graduated with Honors in Computer Science

B.S. Computer Science

• Transferred

UW – Eau Claire

UW – Eau Claire