Zach Hersick

I build machine-learning systems and the software around them.

B.S. Computer Science, University of San Diego. Data Science & AI concentration, graduating May 2027.

Email me GitHub LinkedIn Résumé

Currently: Vision-Guided Follow-Me Robot, Qualcomm Senior CapstoneSeptember 2026 – May 2027

Selected projects

Two deployed applications. Each has a live demo you can try and source you can read.

AnomalyOps Copilot

Python · FastAPI · OpenAI API · PostgreSQL · pgvector

A retrieval-augmented Q&A service over a codebase and its documentation. Every answer cites the exact source lines it came from, and it refuses questions it can’t back up. A read-only operations agent can also pull context from the anomaly API.

Hit@5, up from 40%: 93.3%
MRR@5, up from 0.256: 0.733
indexed chunks from 32 documents: 133

Ingests 32 source documents into 133 chunks using OpenAI embeddings and pgvector.
Gives the agent four read-only anomaly API tools with Pydantic schemas and evidence guardrails.
Raised Hit@5 from 40% to 93.3%, MRR@5 from 0.256 to 0.733, and pass rate from 40% to 75% across 20 evaluation cases.

Try it: Ask “How are alert events grouped and assigned severity?”

Live demo Source on GitHub

Run 1 dashboard with stored anomaly summary metrics and an anomaly type distribution chart.

Industrial Anomaly Detection and Alerting

Python · pandas · scikit-learn · FastAPI · SQLite · Streamlit · Docker

An end-to-end pipeline that generates synthetic industrial sensor data, scores anomalies, groups them into alerts, and serves the results through a FastAPI service and a Streamlit dashboard.

accuracy on a held-out test set: 97.69%
anomaly recall: 94.15%
readings across 10 machines: 50,000

Generates 50,000 readings across 10 machines and engineers 228 time-series features.
A Random Forest reached 97.69% accuracy and 94.15% anomaly recall on a held-out test set.
Serves saved models through feature-validated FastAPI inference with SQLite persistence, Docker Compose, pytest, and GitHub Actions CI.

Try it: Enter Run ID 1 and select “Load Dashboard.”

Live demo Source on GitHub

Now

Vision-Guided Follow-Me Robot

Qualcomm Senior Capstone · In progress, September 2026 – May 2027

An embodied AI system that identifies and follows a selected person or object, using Track Anything on Qualcomm’s Dragonwing IQ9 platform.

Designing the loop from perception to motor control: turning the target’s position and distance into robot motion, with limits for speed, minimum distance, lost targets, and an emergency stop.
Building offline and live-camera tests to measure tracking confidence and find where it fails: occlusion, losing the target, latency, and changing light.

Experience

June 2025 – July 2026

Junior Engineer Intern/Contractor, AI & Python

Syneos Health

Designed reusable Django CMS components for a pre-release training platform.
Built a standalone interactive game system from an initial scaffold, covering backend logic, serialization, and UI integration.
Worked with engineering leadership on architecture, implementation, landing-page design, and debugging across the Python/Django stack.

Education

Expected May 2027

B.S. Computer Science

University of San Diego

Data Science & AI concentration · GPA 3.59/4.0