CVPR Author Analytics
Took a rough university scraping script and rebuilt it into a tested, reusable Python package, working with AI to design the CLI, the analytics, and the offline tests.
AI-Augmented Engineering
How I build production software with AI
I treat AI as a coworker, not a substitute for understanding my own code. It helps me move faster, ship more, and carry what I learn from one project into the next.
The Process
A repeatable, business-first workflow
Every project follows the same path, from the business problem to knowledge I can reuse later.
01
Define the business problem
Start with the real problem. Who it hurts, what it costs, and what "fixed" actually looks like.
02
Document requirements
Write down the scope, constraints, and what success means before any code gets written.
03
Design the architecture
Map out the system, data, and integrations, then poke holes in the design with AI before I commit to it.
04
Generate an implementation plan
Break the work into small, checkable tasks, using Claude Code as a planning partner.
05
Build, test & review
Build, debug, and refactor with AI, then review its output hard before I trust any of it.
06
Document & store knowledge
Write down what worked and why, and save the prompts, decisions, and patterns in Obsidian.
The Toolkit
AI as a coworker, knowledge as an asset
Claude Code & Claude Cowork as a dev partner
I use Claude for planning, building, debugging, and code review, while staying the one who understands and owns every decision.
Obsidian as a structured knowledge base
Notes, system designs, plans, prompts, and lessons, all linked into a searchable second brain I lean on across projects.
A reusable personal dev database
Because I keep past decisions, prompts, and architecture notes organized and reusable, each new project starts further along than the last.
The Toolchain
The AI skills & tools I build with
A practical set of tools I actually use to ship real software, not buzzwords. This is the stack behind every project on this site.
0
AI skills in my toolkit today, and the list keeps growing
🤖 AI development partner
My day-to-day pair for planning, building, debugging, and reviewing on every project.
🧭 Structured engineering skills
A steady spec, plan, build, verify rhythm instead of ad-hoc prompting, so the results hold up.
🕸️ Multi-agent orchestration
Fan large work out to specialized agents in parallel, then adversarially verify their findings before trusting anything.
🎨 AI-generated UI
Generate and refine production UI quickly, then adapt every component to the project's own design system.
🧠 Knowledge & memory
Capture architecture, prompts, and lessons once, then reuse them so each build starts further ahead.
🔌 Tool integration (MCP)
Give AI real, scoped access to the systems a business runs on, and build new integrations when one doesn't exist yet.
Together these make AI a dependable part of how I work: faster builds, fewer regressions, and knowledge that adds up over time.
Knowledge System
My AI-assisted project memory, visualized
A live view of how my projects, skills, and knowledge connect inside my Obsidian knowledge base. Hover a node to see its links, or drag one to give it a nudge.
Proof, not theory
These skills, applied across my projects
Where each part of the toolchain actually showed up in real work.
Dr. Dish Score Tracker
Pulled a scattered internship codebase into a clean Go and PostgreSQL service, working with AI to plan the architecture, write the tests, and ship a live demo dashboard.
Smart Money Signals
Researching a financial foundation model with AI to design and evaluate data-driven trading signals from market data.
This Portfolio Site
Designed and built with an AI partner, including the WebGL hero shader and the animated knowledge graph, through a lot of fast back-and-forth.
AI Knowledge & Workflow System
A reusable second brain. Architecture notes, prompts, and lessons live in Obsidian, connected with knowledge graphs so every project starts ahead.