Projects
The work, unfolded.
Click any project to see the full process — from the first sketch to the shipped product.
01
Automation
Job Application Bot
Python + Playwright automation that applies to 20+ jobs per hour with AI-generated, role-specific cover letters.
View Process +
Try it — click Auto-Apply on any listing
JobSearch — 52 results
🏢
Senior Project Manager
Acme Corp · Austin, TX
📋 Logged → apps_2026.db · cover_letter generated · submitted via Greenhouse
📈
Program Manager II
Vertex Health · Remote
📋 Logged → apps_2026.db · cover_letter generated · submitted via Workday
⚡
Technical PM — AI Products
NovaTech Systems · San Francisco, CA
📋 Logged → apps_2026.db · cover_letter generated · submitted via Lever
🌎
Operations Project Lead
ClearPath Federal · Washington, DC
📋 Logged → apps_2026.db · cover_letter generated · submitted via Greenhouse
How it was built
!
Problem
Job hunting is repetitive work that bottlenecks on human attention
Applying to 50 jobs means filling identical fields across Greenhouse, Workday, and Lever for hours. Human attention is the bottleneck, not qualification. This is a system problem, not a willpower problem.
✎
Sketchbook
Manual research: applied to 30 jobs, documented every form
Manually applied to 30 jobs and catalogued every field, error state, and page structure. Drew a detection flowchart for identifying which ATS is in use before any form filling begins. Workday alone had 7 edge cases.
↳ “Workday has a different modal pattern on mobile — flag this”
◇
System Design
Detect → Fill → Generate → Submit → Log pipeline
Modular Python pipeline: one module detects the ATS, another fills the form fields, a third generates role-specific cover letters via Claude API, and a tracker logs every application outcome to SQLite. Prioritized Greenhouse first (highest volume), then Workday, then Lever based on observed job listing frequency.
PythonPlaywright
Claude APISQLite
✓
Final Product
47 applications tracked, 23 submitted automatically
Working pipeline covering Greenhouse, Workday, and Lever. AI cover letters tailored per role description. Real-time dashboard showing applications found, submitted, and pending. Used in production during an active job search — managed feature prioritization to stay ahead of Workday’s quarterly form structure changes.
02
Game Dev
Hollow Forge
A visual novel mystery built in React — 316-node branching dialogue tree with AI voice acting and custom art.
View Process +
Try it — playable scene
Hollow Forge — The Forge District
THE FORGE — BEFORE SHIFT BELL
NARRATOR
The forge smells like iron shavings and bad decisions. The man behind the counter doesn’t look up.
How it was built
!
Problem
Building a branching story at scale without a proper system
Complex narrative games require hundreds of interlocking choices. Tracking character states, dialogue branches, and item triggers across a 300+ node story without structure leads to spaghetti code and broken playthroughs.
✎
Sketchbook
Hand-mapped dialogue trees and character arc sheets
Started on paper: character motivation maps, scene beat sheets, and a full flowchart of branching choices. Identified 4 character routes and 3 endings per route before writing a single line of code.
↳ “never start coding a VN without the full tree on paper first”
◇
System Design
React component architecture + JSON-driven dialogue engine
Built a custom dialogue engine in React with JSON scene files, flag-based state management, and branching logic. Integrated ElevenLabs for AI voice acting and Dreamina for character portraits generated on demand.
ReactJSX
JSON scene graphElevenLabs API
Dreamina
✓
Final Product
Interactive VN with 316 dialogue nodes, built solo
A fully playable mystery VN with branching choices, character relationship tracking, voiced dialogue, and an explorable dungeon map. Four character routes, each with their own arc, stakes, and resolution.
03
Tools
Solo Leveling RPG Suite
Built full campaign infrastructure from scratch — SaaS character generator, cog-based Discord bot, SQLite backend, and automated weekly game cycle. 60+ users, $15/month, running live.
View Process +
Try it — build your hunter
HunterReg v2.1 — Character Generator
Arise
CHASSIS
Monarch’s Authority — shadows within your domain answer only to you. Soul resource track: max 5 soldiers.
Shadow Extraction
2 PP
Target a defeated enemy. Extract their shadow — they rise as a soldier under your command.
Shadow Step
1 PP
Slip through shadows — move instantly to any point of darkness within near range.
Ruler’s Authority
1 PP
When an object moves toward you — halt it mid-air and redirect at will. No action required.
Monarch’s Domain
2 PP
Passive: shadows in your zone gain +Rank to all attacks. Active: expand domain — flood the field with your darkness.
Sovereign’s Wrath
3 PP
Once per gate — unleash all soldiers simultaneously. Each strikes for full damage. Field is cleared on resolution.
How it was built
!
Problem
Running a paid PBP campaign needs infrastructure off-the-shelf bots can’t provide
Paid Discord campaigns require character creation, XP tracking, gate sign-ups, loot logging, and turn reminders all working together. Generic bots solve one piece. The campaign needed a system built for its specific workflow.
✎
Sketchbook
Player journey map and full Discord server architecture
Mapped the complete player journey: sign-up → create character → join gate → earn XP → rank up → run trials. Drew the channel structure and listed every bot command needed before writing anything.
↳ “reaction-based sign-ups beat slash commands for new players”
◇
System Design
HunterReg SaaS + discord.py cog-based bot architecture
Built HunterReg as a standalone character generator SaaS. Discord bot with cog modules: gates.py handles sign-ups, hunters.py tracks XP, reminders.py sends turn nudges, schedule.py automates weekly posts. SQLite for campaign state.
Pythondiscord.py
SQLiteReact (SaaS)
✓
Final Product
Live paid campaign at $15/mo — full weekly automation cycle running
Live paid campaign with reaction-based gate sign-ups, automated turn reminders, GM prep pipeline, XP tracking, rank-up trials, and session recap publishing. Full weekly cycle Monday through Saturday. Managed ongoing player onboarding, balance feedback, and weekly scheduling across time zones.
04
Design System
Resume Automation System
One data source → three distinct resume variants with metrics dashboards, case studies, and coordinated cover letters.
View Process +
Try it — generate your variants
resume_builder.py — output generator
V1 — CORPORATE
Lamont Clay
Project Manager
$300K budget managed
V2 — CREATIVE
Lamont Clay
Project Manager — Creative
$300K budget managed
V3 — EXECUTIVE
Lamont Clay
Project Manager — Executive
$300K budget managed
How it was built
!
Problem
Different roles need fundamentally different resume formats
A corporate PM role wants a clean ATS document. A creative agency wants designed layouts. An executive role wants metrics front and center. Maintaining three versions manually is error-prone and breaks the moment one piece of data changes.
✎
Sketchbook
Layout sketches for three distinct design directions
Sketched three resume personas: V1 Corporate (clean blue/white, ATS-first), V2 Creative (teal accent, visual hierarchy), V3 Executive (bold metrics dashboard, embedded case study). Each had its own grid before any code.
↳ “FEMA work → $300K budget + 97% QC rating are the lead metrics”
◇
System Design
Single YAML data source → template engine → 3 PDFs
Python pipeline with a shared data model. Each variant has its own template module reading the same source. ReportLab renders PDFs, python-docx generates Word versions. Cover letters auto-generate from the same data object.
PythonReportLab
python-docxNode.js
✓
Final Product
Three polished variants from a single command
Run one script: get three PDFs, three Word docs, and three cover letters. Update your data once and everything regenerates. Reduced resume update time from hours to under 5 minutes. Plugged directly into the Job Bot pipeline — each of the 23 automated applications used a variant generated from this system, cover letter included.
05
Mobile App
Q-Pathos
Gay male dating app built on a blind chat → face reveal mechanic — React Native + Expo + Supabase, 14-table schema with full RLS, dual brand A/B test, Malta prototype in 10 days.
View Process +
Product overview — swipe through the app
Q-Pathos — Brand · Mechanic · Features
How it was built
!
Problem
Gay dating apps optimize for appearance-first — creating hookup culture even when users want connection
Every major gay dating app surfaces photos before anything else. Users self-select into surface-level interactions because the interface rewards it. The goal was to invert that loop: force conversation to happen first, let the reveal be earned, and make the photo moment feel like something — without moralizing about it.
✎
Sketchbook
Reveal rules, quality filtering, and a 14-table privacy-first schema
Designed the reveal mechanic: unlock after 15 qualifying messages OR 24 hours, whichever comes first — both users need 5+ qualifying messages to trigger. Quality filter defined server-side: emoji-only or short replies don’t count, 10+ words required. No race field in qualifiers (by design). Post-reveal images blurred by default, tap to consent. Two brand directions designed for Malta A/B: Warm Dusk (dark, intimate, Playfair Display) and Clean Signal (editorial, light, Syne).
↳ “The reveal has to feel like something — that’s the whole product”
◇
System Design
One codebase (iOS/Android/Web), Supabase Realtime, server-side quality scoring and brand assignment
React Native + Expo for cross-platform from day one. Supabase handles auth, Postgres, realtime subscriptions, and RLS — look ratings are completely private (no joins expose them), chat threads are participant-only. Message quality scoring runs as a before-insert Postgres trigger. Brand direction (A/B) is assigned 50/50 server-side before user insert so the client never chooses. Token storage uses SecureStore on native and localStorage on web, platform-detected at init.
React NativeExpo
SupabaseTypeScript
PostgreSQLRLS
✓
Final Product
Malta prototype in 10 days — auth, onboarding, blind chat, dual brand, 50-user test
Sprint 1 delivered: full auth flow, conversational onboarding (5-step signal chips + 2 qualifier dealbreakers), blind chat with quality-scored messages, dual brand system live for A/B, and all 14 schema tables with RLS enforced. Testing in Malta with 50 users. Reveal animation — the highest-emotion moment in the product — is the Sprint 2 priority.
06
UX Redesign
APF RBT Training 2.0
Full UX audit and visual redesign of 20 e-learning modules for Autism Partnership Foundation — new design system, two color variants, and a client pitch deck backed by before/after evidence.
View Process +
Try it — select an answer to see the difference
APF_RBT_Training_2.0 — Quiz Feedback Comparison
01
Module Header
■ Before — Plain & White
APF RBT Training 2.0 › Section 1
Module 1.01
ABA: History, Features, & Purpose
This module introduces the foundations of Applied Behavior Analysis — where it came from and the seven dimensions that define it.
■ After — Design 2 Bold
Module 1.01
ABA: History, Features, & Purpose
This module introduces the foundations of Applied Behavior Analysis — where it came from and the seven dimensions that define it.
02
Quiz Feedback — click an answer
■ Before — No Explanation
Question 3 of 8 · Module 1.02
A client puts toys away after "clean up" because this was followed by earning free time. This is an example of:
✗ Incorrect — correct answer is B.
✓ Correct!
↻ Try again
■ After — BCCBA Explanation
Question 3 of 8 · Module 1.02
A client puts toys away after "clean up" because this was followed by earning free time. This is an example of:
BCCBA
Why this answer is correct
Positive reinforcement occurs when a stimulus is added following a behavior, increasing the future probability of that behavior. Free time (an appetitive stimulus) was delivered after "clean up" — making it more likely to recur. This differs from negative reinforcement, which involves the removal of an aversive stimulus, not the addition of a desirable one.
BACB Task List 5th Ed. · A-01 · Cooper, Heron & Heward (2020)
↻ Try again
03
Key Concept Callout
■ Before — Unstyled Box
⚠ Key Concept
The seven dimensions of ABA — Applied, Behavioral, Analytic, Technological, Conceptually Systematic, Effective, and Generality — are the foundation of every behavior-analytic intervention. These dimensions were defined by Baer, Wolf, and Risley in 1968 and remain central to the BACB Task List today.
■ After — Structured Card
◆ Key Concept
RBT Task A-01
The seven dimensions of ABA — Applied, Behavioral, Analytic, Technological, Conceptually Systematic, Effective, and Generality — are the foundation of every behavior-analytic intervention. These dimensions were defined by Baer, Wolf, and Risley in 1968 and remain central to the BACB Task List today.
▶ Remember: BATTEEG is a common mnemonic for the seven dimensions.
#9A031E
Crimson
#2D3142
Navy
#BAB700
Chartreuse
#407B66
Teal
#EFE78F
Quiz Tint
20
Modules redesigned
2
Full design variants
40
Deliverable HTML files
How it was built
!
Problem
20 training modules with no design system and zero visual hierarchy
APF's existing RBT training used dark gradient headers, inconsistent color use (orange callouts, red wrong-answer indicators that confused learners), and no cohesive type scale across 20 modules. The visual language undermined the content's authority and made navigation harder than it needed to be.
✎
Sketchbook
UX audit, two color directions, and a full header rethink
Audited all 20 modules for pattern inconsistencies: wrong-answer red borders (cognitive conflict with error state conventions), orange callout colors with no semantic meaning, flat dark headers with no information hierarchy. Developed two distinct color systems — Design 1 (earthy neutral) and Design 2 Bold (crimson / navy / chartreuse) — with a documented decision rationale for each UX choice.
↳ “Red outlines on wrong answers fight the user's instincts — removed entirely”
◇
System Design
CSS custom properties token system applied across all 20 modules via automated script
Built the full design language using CSS custom properties (‘:root’ tokens) so all color decisions cascade consistently. Designed the bubble header — a chartreuse oval with navy corner triangles, red-tinted description pill, and center-aligned breadcrumb hierarchy. Wrote a Python transformation script to apply the full design system to all 20 module prototypes automatically, detecting each module's unique header content and rebuilding it to spec.
HTML / CSSCSS Custom Properties
PythonUX Audit
Design Systems
✓
Final Product
40 deliverable files across 2 design variants + client pitch deck with $15K proposal
Delivered 20× Design 1 and 20× Design 2 Bold HTML prototypes — each a fully interactive single-file module with micro-lesson carousel, quiz engine, key concepts, and RBT callouts. Paired with a before/after audit report and a client pitch deck including a 4-color palette rationale, evidence-backed redesign case, and a $15K implementation proposal for APF's full rollout.