Junior Full-Stack • Automation • Data Tools

Ryan MacKellar

Entry-Level Full-Stack Developer building operations software, workflow automation, and data-heavy tools.

Applied Mathematics graduate with a private quote-to-payment platform for WrenchTime Cycles and a Python Excel extractor for industrial quoting data.

WrenchTime OS: quote-to-paymentHNI: Python Excel extractorApplied math, May 2026Next.js / Python / PostgreSQL
View Projects Download Resume
GitHubLinkedInEmail
Projects

Real systems before learning projects.

WrenchTime shows product ownership, HNI shows reviewable automation code, and the AI project supports math and machine-learning depth.

Quote-to-Payment Operations System for a Real Service Business

WrenchTime OS

Private full-stack platform built to support quote intake, scheduling, payments, expense review, attribution, and advisory workflows for an active service business.

System depth

Request, review, quote, deposit, scheduling, job, closeout

Production shape

Auth, Stripe, email, uploads, calendar, database models

Risk boundary

AI and ad recommendations stay advisory and review-gated

  • Built quote-management flows from inquiry through scheduling, deposits, final balances, expenses, and attribution.
  • Integrated protected admin access, Stripe payments, calendar scheduling, email, uploads, and PostgreSQL-backed data models.
  • Kept AI advisory workflows read-only and settings-controlled instead of automating customer or ad-account actions.
Next.jsReactTypeScriptPostgreSQLPrismaSupabaseVercelStripeAuth.jsResendUploadThingGoogle CalendarOpenAI API

Private-source system; the public case study documents architecture and scope without exposing business data.

Review signal: Review the case study for architecture and scope; the business site shows the operating context.

Case StudyBusiness Site
WrenchTime OS dashboard showing request, scheduling, scheduled job, and completed job queues.

Command center

Operational dashboard

The dashboard exposes queue health, open work, and next-step decisions instead of presenting a static admin table.

Problem

A service business creates state fast.

Customer requests, quote assumptions, approvals, deposits, appointment windows, invoices, expenses, and follow-up all need a shared source of truth.

Build

Modeled the operation as a product workflow.

The admin dashboard, customer request flow, quote approval, payment handoff, mobile view, and growth intelligence views all map to one operating lifecycle.

Result

Recruiter signal beyond CRUD screens.

The work demonstrates product judgment, data modeling, integration boundaries, responsive UI, and operational thinking inside one coherent system.

Visual evidence

What the work proves

Each image is paired with the specific technical signal it supports.

WrenchTime OS lifecycle diagram from request through growth intelligence and bookkeeping readiness.

System model

Lifecycle architecture

A full request-to-growth map shows the product thinking behind the screens and the audit trail each stage creates.

WrenchTime OS review request screen with quote sending controls and communication history.

Admin workflow

Quote review and communication

Quote links, reminder controls, and customer communication sit in the same review surface so the next action is obvious.

WrenchTime OS growth intelligence overview with trends, lead funnel, and tracking health cards.

Growth intelligence

Operating metrics with guardrails

Revenue, lead quality, spend, and tracking health are presented as decision support rather than unchecked automation.

Expand full WrenchTime evidence trail+
WrenchTime Cycles customer request service form with guided service categories.

Customer intake

Guided request flow

The customer-facing request page gathers structured service context and photos before an admin ever opens the queue.

WrenchTime customer quote review page with approved status, fee breakdown, and deposit action.

Customer handoff

Quote approval and deposit path

The public quote view turns an estimate into a clear approval, deposit, and scheduling handoff.

WrenchTime OS expense management page showing year-to-date totals, filters, and transaction review.

Finance ops

Expense management

Bank transaction review, manual expense records, receipt status, and year-end export are part of the same operating system.

WrenchTime OS mobile dashboard showing navigation, review requests button, and queue summary cards.

Responsive surface

Mobile admin

The queue remains usable on a phone, which matters for a business that moves between shop, customer, and field contexts.

WrenchTime OS AI recommendation review with evidence, instructions, and approval controls.

AI boundary

Review-gated recommendations

Recommendations include evidence, confidence, manual instructions, and approve/reject controls before any operator action.

Recruiter takeaway: This is the strongest hiring signal because it combines real operational ownership with production-style integration work and careful AI boundaries.
Industry Capstone / Data Automation

Hope Network Industries Data Extraction Tool

Python desktop automation tool for turning batches of inconsistent quote spreadsheets into one standardized Excel workbook.

Input problem

Inconsistent quote workbooks and source cell locations

Core design

Editable Excel mapping layer drives reusable Python logic

Output contract

One clean row per package with row-level errors

  • Served as primary coder and data compiler on a university-sponsored applied mathematics capstone.
  • Built a configurable Excel mapping layer so extraction rules could change without editing Python.
  • Implemented reusable parsing functions for direct reads, fallbacks, conditional counts, sums, and filename checks.
PythonCustomTkinteropenpyxlpandasPyInstallerExcelData Automation

Public materials keep NDA-covered company IP and source documents out of view.

Review signal: Review the source and README in the extractor folder; this is the clearest public automation code sample.

Source + README
HNI extractor problem overview showing messy synthetic inputs, configurable extraction, and clean output.

Problem to output

Normalized extraction flow

The overview compresses the whole automation story: messy synthetic inputs, configurable extraction, and standardized output.

Problem

Spreadsheet variation was the real enemy.

The task was not just reading Excel cells. It was normalizing different workbook layouts without hard-coding every source document.

Build

Moved business rules out of Python.

A mapping workbook defines candidate cells, ranges, fallback rules, keyword checks, counts, sums, and conditional functions the engine can evaluate.

Result

A reviewable automation artifact.

The public reconstruction keeps sensitive data out while still proving the architecture, workflow, function set, and clean output contract.

Visual evidence

What the work proves

Each image is paired with the specific technical signal it supports.

HNI mapping template flow showing source sheets, mapping rules, standardized output, and rule evaluation.

Rule layer

Mapping template flow

Business users can adjust extraction behavior in an Excel mapping row while the Python engine stays reusable.

HNI custom functions table listing extraction functions and what each proves.

Function language

Custom extraction functions

Fallbacks, counts, sums, filename checks, keyword searches, and paired row checks are documented as reusable commands.

HNI clean output example showing standardized spreadsheet columns and technical proof points.

Output contract

Clean workbook preview

The evidence lands in a simple analysis-ready table: package file, quote ID, category, counts, hours, flags, and checks.

Expand HNI implementation evidence+
HNI extraction workflow screen with mapping file, packages folder, output file, and status log.

Workflow

Extraction run reconstruction

The desktop workflow shows mapping selection, package selection, output naming, and a readable status log.

HNI project architecture diagram showing CustomTkinter UI, mapping workbook, engine, readers, and compiled workbook.

Architecture

Module map

The architecture diagram separates UI, mapping workbook, extraction engine, quote readers, and compiled workbook export.

Recruiter takeaway: This case study is the clearest public code sample: it shows Python automation, spreadsheet parsing, user-configurable rules, and careful NDA-safe documentation.
Supporting ML / Applied Math Project

AI From First Principles + Motorcycle Part Detector

Honors project exploring machine-learning foundations through written analysis, Python notebooks, and an applied motorcycle-part detector.

Math path

ML foundations, MNIST notebooks, NumPy-level reasoning

Applied model

YOLOv8 detector trained on motorcycle-part imagery

Hiring role

Supporting evidence for AI curiosity and technical range

  • Implemented MNIST experiments at decreasing abstraction levels, from high-level models toward manual NumPy work.
  • Trained a YOLOv8 computer-vision model on personal motorcycle-part imagery.
  • Documented the learning path so the work shows reasoning depth, not just model usage.
PythonMachine LearningComputer VisionNumPypandasMatplotlib

Review signal: Best used as supporting evidence for math and ML curiosity, not as the primary hiring signal.

GitHub
Motorcycle part detector output showing labels for clutch lever, carburetor, discs, chain, and sprockets.

Computer vision

Full-bike part detection

The detector identifies multiple visible motorcycle components in one frame, including controls, brakes, chain, and sprockets.

Problem

Move from theory to observed predictions.

The project pairs first-principles machine-learning study with a small computer-vision model that produces inspectable outputs.

Build

Trained on owned, domain-specific imagery.

Motorcycle photos create a concrete labeling problem: the model must find visible components across full-bike and close-up views.

Result

A compact applied-ML proof point.

This supports the portfolio's applied math story while keeping WrenchTime and HNI as the main software and automation signals.

Visual evidence

What the work proves

Each image is paired with the specific technical signal it supports.

Motorcycle detector output showing taillight, rear brake caliper, and rear disc labels.

Close-up validation

Rear brake and taillight view

Close-up detections test whether the model can still recognize parts when the full motorcycle context is cropped away.

Motorcycle detector output showing front disc and front brake caliper labels.

Part detail

Front disc and caliper

The detector localizes brake hardware in a tighter, lower-angle view where component boundaries are more specific.

Motorcycle detector output showing rear brake caliper and rear disc labels from an alternate angle.

Alternate angle

Rear disc and caliper

The fourth output adds angle variation, helping show this is a model behavior review rather than a single lucky screenshot.

Recruiter takeaway: Use this project as supporting proof of mathematical curiosity, model literacy, and the ability to explain AI work with concrete visual artifacts.
Skills

A toolkit for software, data, and practical engineering.

Software

TypeScriptReactNext.jsHTMLCSSGitGitHubVS CodeLinux

Backend/Data

PostgreSQLPrismaSupabaseAPIsserver actionsroute handlers

AI/Data

PythonNumPypandasMatplotlibOpenAI APImachine learning fundamentalsstatistical modelingExcel automationopenpyxl

Engineering

SolidWorksCADArduinoRaspberry Pioscilloscopes3D printingCNC routing
Experience

Operator context, service work, and technical teaching.

Founder / Developer

WrenchTime Cycles / WrenchTime OS

Built and operated quote, scheduling, payment, analytics, and automation workflows for a real service business.

Server

Tupelo Honey

Worked in a fast-paced, guest-facing environment requiring clear communication, prioritization, and reliable execution under pressure.

Physics Tutor

Grand Valley State University

Helped students reason through physics concepts, problem-solving strategy, and quantitative thinking.

Education

Applied mathematics with engineering sciences context.

Grand Valley State University

Bachelor of Science in Applied Mathematics
Minor: Engineering Sciences
Graduated May 2026
GPA: 3.914 / 4.000
Summa Cum Laude
Frederik Meijer Honors College
Dean's List Every Semester
Certification

Certified SolidWorks Associate

Grand Valley State University. Issued Apr 2023. Credential ID: C-3QN2AXB79R.

Resume

Fast links for recruiters and hiring teams.

Resume, LinkedIn, and GitHub are grouped here for quick evaluation after the project work has done the talking.

Download ResumeView LinkedInView GitHub
Contact

Open to junior full-stack, automation developer, technical product builder, and startup generalist roles.

Email

ryanmackellar7611@gmail.com

Open ->

GitHub

github.com/mackelry-hub

Open ->

LinkedIn

www.linkedin.com/in/ryanmackellar

Open ->

WrenchTime

www.wrenchtimecycles.com

Open ->