01 / Define
Establish the behavior the check is intended to detect and place it within the regression workflow.
Architecture / automation / human review
Selected work · 2026ELTON COSPER
Regression architecture,
built to be checked.
My lead work is an ongoing AI-assisted regression system. I own its architecture, infrastructure, reusable skills and workflows, and personally validate the Playwright checks in a headed browser.
How this works
The lettering is HTML text with a WebGL texture drawn inside a matching letter mask. The smoke is generated with code, not a video.
- Broad flow: layered fractal noise establishes the large, slow-moving shapes.
- Curled folds: a curl field bends that flow into smaller swirls.
- Fine variation: a finer noise layer gives the purple ink depth without particles or glow.
A pointer adds a very small local disturbance. Motion pauses outside the viewport or in a background tab. Reduced-motion and unsupported-graphics settings use static lettering.
AI-assisted implementation under my creative direction and evaluation.
Static lettering is available without animated graphics.
LEAD CASE STUDY / regression-architecture
The system around
the test matters.
Ongoing professional work
I architect and build an AI-assisted regression workflow, from its infrastructure and reusable instructions to the browser checks I validate myself.
My contribution is the architecture of the working system: deciding how its components fit together, creating the infrastructure, and organizing the commands, skills, workflows and metadata that support testing. Playwright scripts are one part of that system.
I use AI to help create and repair scripts. I own the structure, guide the work, inspect the output and personally validate behavior in a headed browser. That is system ownership with AI assistance; it is not a claim that I handwrite every script.
Infrastructure and organization
I built the supporting infrastructure and organized workflow metadata around the regression work. My rationale is to keep test intent and the work needed to run, inspect and maintain it connected, instead of treating scripts as isolated files.
Reusable skills and commands
I created skills, commands and workflows to make the testing approach reusable. They give AI-assisted work a defined process that I can inspect and refine as the system develops.
Browser automation and repair
I direct AI-assisted Playwright creation and repair, then inspect behavior in an actual headed browser. A generated script needs to demonstrate the intended check; plausible code alone is not the result.
Tool selection and integration
I identified and installed external MCP tools, including Playwright MCP for browser automation and Context7 for current documentation. Selecting and integrating those tools is part of my developer-workflow ownership. Documentation informs implementation; behavior still needs testing.
REGRESSION / VALIDATION LIFECYCLE
02 / Build and inspect
Create or repair the Playwright script with AI assistance, then personally inspect its headed-browser execution.
03 / Challenge
Deliberately introduce a fault and check that the test fails for the expected reason. A passing run by itself does not show whether a test can detect a regression.
04 / Restore and recheck
Restore the intended behavior and rerun the check. Review unexpected results and refine the script or workflow as needed.
This work is ongoing. The case study describes my contribution and approach without identifying the organization or product. Proprietary code, internal links and test identifiers are not included. Quantitative performance and adoption results are not claimed.
01 / projects
Open a project.
See where it stands.
Each file separates existing work from the next step. Prototypes and starter labs are labeled plainly.
7 projects / Select a project to see its current scope.
Regression architecture / browser validation
AI-assisted regression system
Architecture, infrastructure, reusable skills and commands, AI-assisted Playwright scripts, and personally validated browser checks.
Read the technical case studyAgent chaining / delivery review
Front Door Web Studio
A website delivery workflow with bounded agent roles, independent checks and human approval.
Writing tools / content QA
Inkwright Taste Engine
Paragraph-level checks for word choice and style patterns. The larger writer IDE and BYOK support are planned.
Research / source health
AI Librarian
Research collection and processing reports, with an arXiv focus and visible source-health checks.
Editorial workflow / publication checks
The Morning Commit
An editorial process with source review and private-data checks. The AI Librarian handoff is planned.
Technical answers / evaluation
IBM TechQA
A technical question-answering lab direction. Evaluation cases and measured results are still to come.
API onboarding / training
Northstar API Enablement Lab
An API enablement lab direction focused on practical onboarding and repeatable response checks.
02 / review-process
A handoff needs
something to check.
Front Door Web Studio is my working prototype for supervised agent delivery. The brief travels with the work, and a person makes the final decision.
FRONT DOOR / INTENDED REVIEW CHAIN
Set the brief
Goal, source material, acceptance criteria and approval boundaries.
Bound the work
Research, design and implementation each receive a specific task.
Inspect the evidence
A fresh review compares the result with the brief and sends failures back.
Keep a human gate
I accept the result or request a correction. Publication needs approval.
└─ Correction loop: build / review / revise / review again
A process model, not a claim that a delivery is running.
03 / research-to-editorial
Two projects.
One planned connection.
A research report could become an input to an editorial board. That handoff still needs to be built and verified.
In development
AI Librarian
Source-health and processing reports, with arXiv research collection as a future direction.
[ PLANNED HANDOFF: RESEARCH REPORT ]
In development
The Morning Commit
Editorial selection, drafting, source review and human approval before publication.
The connection is planned. These projects are not presented as a completed end-to-end system.
04 / about-elton
I start with
the user's problem.
I approach AI work by defining what should happen, inspecting where it fails and explaining the result to the people who need it.
I'm focused on AI implementation and enablement. My ongoing professional regression work and independent projects show the decisions, checks and remaining work behind that practice.
05 / contactLet's make the
Let's make the
next handoff clearer.
I'm interested in AI implementation and enablement roles where practical guidance and human review matter.
Connect through my verified LinkedIn profile.