Data analysisWhen the biggest count tells the wrong story.Explore
Independent analytics study / Synthetic data
Count the defects. Then question the count.
Which production cell deserves investigation first? A high reject count may simply reflect a
larger workload. Choose a time window, make a prediction, and compare counts with
volume-adjusted rates.
Created for this portfolio with AI assistance. All 36 records are invented; this is a method
demonstration, not an employer result.
Weeks 1–12 · 36 synthetic records
Cell A 2.05%
2.05%
Cell B 3.00%
3.00%
Cell C 4.60%
4.60%
Bars share a 0–6% scale.
Cell C has the highest reject rate. Cell A has the largest reject count. Volume changes the
priority.
Inspect the data and reasoning
The calculation
Reject rate (%) = total rejected units / total inspected units × 100. Aggregate the counts
first so weeks with different volumes receive the correct weight.
Synthetic totals / Weeks 1–12
Cell
Inspected
Rejected
Rate
A
12,600
258
2.05%
B
6,600
198
3.00%
C
3,000
138
4.60%
What this does not establish
A higher rate is a signal to investigate, not a cause or a performance verdict. Product mix,
inspection practices, and sample size could explain differences. A real study would validate
those factors and examine uncertainty before recommending a change.
Reproduce it
The downloads reproduce the full 12-week study. The controls above filter that same dataset
in your browser. Download these three files into one folder and run python3 quality-analysis.py. Python’s standard-library SQLite engine runs the supplied SQL. Validation checks reject
counts, positive volumes, and duplicate cell-week keys.
Agentic systemsInspect the checks before trusting the answer.Explore
Interactive lab / Supervised agent workflow
Show the work. Earn the next step.
An agent needs a plan, usable inputs, and a check before its answer reaches a person. Try the
happy path, then give this small workflow a broken reading.
An independent simulation built for this portfolio with AI assistance. These light readings are
synthetic. It runs entirely in your browser, with no live AI model or external tool calls.
The task
Calculate the mean of four sample light readings. Report any excluded input.
Synthetic light readings / lux
Reading
Value
Use
R01
180
Pending
R02
200
Pending
R03
220
Pending
R04
240
Pending
A method demonstration. No lighting or safety recommendation is made.
01 Plan
Define the calculation and its checks.
02 Read the sample
Load the built-in records.
03 Validate
Require finite, nonnegative numbers before calculating.
04 Calculate
Show the arithmetic and excluded records.
05 Human review
The result waits for your approval.
Choose an input, then plan the workflow.
01 / DURA · Author & engineer
Progress should travel with you.
An open-source learning platform, built for the messy reality of learning: interrupted
sessions, unreliable connections, and coming back tomorrow.
I built the curriculum and the platform beneath it, including offline progress, spaced
repetition, and in-browser code execution.
Finish a lesson while offline, reconnect, then repeat the sync. Can older server data
erase your newer progress?
Learn offline
Reconnect
Sync again
This device3 / 5
Server2 / 5
Illustrative sample: keep the higher progress value when reconnecting.
Why this approach?
DURA uses monotonic merging: completion is combined with OR; accumulated time and
scores use the maximum. This simplified model demonstrates the maximum rule. It is not
a connection to a learner’s account.
Inspect the engineering evidence
Lesson IDs include their phase and module so two lessons numbered “01” cannot share
progress by accident. A regression test completes one lesson, opens matching numbers
elsewhere, and checks that completion stays with the original lesson.
I didn’t discover problem-solving when I started writing code.
Precision manufacturing taught me to investigate the process, question assumptions,
and verify the result. I’ve brought that discipline to my work at Rolls-Royce MTU
since 2021.
Alongside the job, I completed Coding Temple’s software program and began building
systems of my own.
Software is the next natural step: the same care, applied to useful products,
trustworthy data, and AI-assisted work I can stand behind.
2019–2020Colt Atlantic Services / IMTSolving problems where they happenExplore
Industrial field technician · training program
The work
Supported industrial leak repair and assembly work, including flange clamps, valve seats, and line enclosures. Completed job documentation and integrity checks while traveling to plant sites.
What carries forward
Observe the actual conditions, understand the constraints, and verify the repair. Those habits still guide my debugging.
Follow the connection
From investigating a repair to debugging
The public DURA story follows a software failure, the decision behind a fix, and evidence a reader can inspect.
2020–2021Keselowski Advanced ManufacturingBuilding in metal, layer by layerExplore
Additive Manufacturing Technician
The work
Operated GE, SLM, and EOS metal additive equipment, supported aerospace assembly, and inspected parts. Handled materials, finishing, and production travelers.
What carries forward
A design becomes useful through a repeatable process. Material, tooling, inspection, and documentation all shape the result.
Follow the connection
From a repeatable process to a reproducible analysis
The synthetic study includes its inputs and analysis files so a reader can check the result. It uses no production records.
Joined in December 2021. Machine engine components on HELLER centers with Siemens controls, interpret control plans, adjust setups, and inspect finished work. Moved into the floater role in August 2026.
What carries forward
Read the specification. Investigate the process. Verify the output. That discipline is the foundation of how I approach software.
Follow the connection
From inspection to asking better questions
My independent quality study lets you test the difference between a defect count and a defect rate. The records are synthetic.
2023–2025Coding TempleTurning curiosity into software skillsExplore
Software engineering program
The work
Studied software from December 2023 through June 2025 alongside full-time manufacturing work, completing a Software Engineering Certificate and building full-stack coursework projects.
What carries forward
The medium changed, but the process felt familiar: understand requirements, build a working system, and test what it does.
Follow the connection
From coursework to a public product
DURA is a place to put software learning into practice. Its public walkthrough gives visitors an entry point into the product.
TodayDURA / Durwood StudiosBuilding a place to keep learningExplore
Author, founder & software engineer
The work
Built an open-source learning platform with offline progress, spaced repetition, and in-browser code execution. Established Durwood Studios LLC in 2026.
What carries forward
Platform engineering brings the pieces together: product design, persistent data, application code, and verification.
April 23, 2026Microsoft & SimplilearnExtending the toolkit into AIExplore
Microsoft AI Engineer Program
The work
Completed the program in April 2026, including capstone work and study in Python, machine learning, deep learning, generative AI, Azure AI, and agent workflows.
What carries forward
I use AI-assisted development with written governance and human verification, and build MCP tools that let agents interact with software.
Follow the connection
From AI study to inspectable practice
A public DURA story shows how I make AI-assisted development traceable and keep human review part of the process.
05 / Program completion Microsoft AI Engineer ProgramSimplilearn · Microsoft program
Explore credential
Issued April 23, 2026
Certificate of successful program completion issued by Simplilearn. This is a program completion certificate, distinct from a Microsoft role-based certification.
A learner’s progress was on their device. A network failure still made it inaccessible.
Cloud unavailable
Local identity retained
Learning continues
Open the engineering story
The decision
I changed startup to resolve the locally cached identity before choosing the encryption key, so a cloud outage would not be mistaken for signing out. The learning dashboard also gained a usable fallback.
The evidence
The implementation and commit explain the original failure, the local-session fallback, and the decision to reserve key removal for an explicit sign-out.
AI-assisted implementation; authored and integrated under my direction.
Using AI to write code is easy. Leaving a reviewable record of how it was used takes engineering.
Protected path changed
Provenance declared
Ready for review
Open the engineering story
The decision
I maintain a provenance check that reads protected paths from CODEOWNERS and inspects commit messages for an AI-assisted or explicit human-only declaration. I updated it to support the project’s newer trailer format.
The evidence
The checker is executable repository code. It helps enforce a declared workflow; a provenance label is not proof that a change is correct. Testing and human review still carry that responsibility.
A concrete example of how I direct AI-assisted work rather than treating generated output as finished work.
