Occidental College · CS and Japanese Studies · class of 2027

I build the automation. Then I go find where it's wrong.

Across content pipelines, research coding, price models, and security reporting, the pattern is the same: building the automation is the easy half. Proving the output holds is the job.

Email me about a role Read the code

or copy it: bng@oxy.edu

Four automations, each checked against the work it replaced

The range looks scattered on a resume. It isn't. It's one move applied to different work: take something a person did by judgment, build a pipeline for it, then go find out whether the pipeline actually holds.

Coding lecture video Five trained coders scored this by hand. My pipeline agrees with the panel 93% of the time.
Writing SEO pages A marketer wrote these one at a time. Mine ships from a single keyword and regenerates any draft that fails its own editorial gate.
Forecasting prices I built the model, then found the temporal leakage inflating my own results and corrected it.
Security reporting Domain, firewall, and WordPress assessments, still done by hand at work. The next thing I want to automate.

The automation is never the interesting part. Proving the output survives someone checking it is.

Research that held up to an expert panel

VISTA: Dense Multi-Label Classroom Coding with Vision‑Language Models

Second author · DataMFM Workshop, CVPR 2026 · DataMFM‑30

A vision-language pipeline that reads raw lecture video and produces COPUS activity codes without a human in the loop. The hard part was never the model. It was proving the output could be trusted against people who do this professionally. Built with three other engineers on a shared codebase and run unattended across the full corpus, which is a different exercise from a notebook that works once.

Video processed unattended: 4 TB
Agreement with expert panel: 93%
Engineers on the codebase: 4

Work that shipped, with the numbers to show it

91/100 RankMath, on a draft it wrote and scored itself

Shopify Blog AI Pipeline

One keyword in, a review-ready Shopify draft out. Claude writes the copy under a strict output schema, Gemini generates and self-QAs the images, and a deterministic SEO/AEO gate scores every draft against hard minimums before anything ships. Publishing is draft-only by design: the pipeline creates an unpublished draft over the Shopify GraphQL Admin API and a human clicks publish. Built for a live store, with least-privilege scopes, isolated secrets, and credential-leak hooks.

Python · Claude · Gemini · Shopify GraphQL · MCP · pytest

10 yr of price history, and one leakage bug in my own model

Multi‑Stock Prediction

LSTM and MLP forecasters testing whether correlated-stock features improve single-stock prediction. The result that mattered was negative: I found temporal data leakage inflating my own performance numbers and corrected it. The honest model is worse and real.

PyTorch · time-series · rolling-window validation

$1,000 flat per site, hosting and local SEO included

ngdigital.org

A web design practice I run for local businesses in the East Valley. The site gets built and shown before anyone pays. Real clients, real invoices, and the reason the content pipeline above exists at all.

Design · build · deploy · client management

Now

124 blog posts rebuilt to SEO 80+

Win‑Pro Consultancy

Singapore · Digital & E‑Commerce Intern and Cyber Security Intern · May 2026 – present

Two roles in one seat, and most of my week goes to the first. On the digital side: audited and rebuilt 124 blog posts to SEO scores of 80 or better, ran the company's Shopify storefront end to end (listings, quotations, order processing), and prototyped automation for the order workflow. On the security side: a web-application penetration test, an internal-network VAPT across two client engagements, and the formal vulnerability assessment reports that came out of both. Working toward the PNPT.

3.93 GPA · Dean's List 2024–26

Occidental College

Los Angeles · BA Computer Science and Japanese Studies · May 2027

Coursework in data structures, algorithms, databases, software engineering, data science, and mathematics for artificial intelligence.

Looking for new‑grad roles starting 2027.

Automation, applied AI, software, security, or quantitative research: I read all of it. If you want the long version, the code is public and the paper is real.

bng@oxy.edu github.com/bng11299 LinkedIn