# PROJECTS — one block per project, shown as filterable cards.
# `domain:` lines are the project's domain tags — they show on the card and
# become filter chips at the top. Repeat the key for multiple domains.
# Media panel: `image: assets/x.png = alt` for a screenshot, `video: assets/demo.mp4`
# or a YouTube URL for a demo, or `placeholder: hint text` for a striped slot.
# Links: `button:` = solid primary, `link:` = outlined, `note:` = plain grey text.
template: projects
label: PROJECTS
title: Selected work

Eight systems across product, vision, research and embedded. Pick one to see it up close.

---
title: Setsquare
domain: Product
domain: Full-Stack
badge: LIVE · SOLO BUILD
badgeHighlight: true
stat: 22 = tools in one suite
stat: solo = design and build
stat: free = no signup to use
button: Open Setsquare -> https://setsquare.pages.dev
image: assets/Projects/setsquare.png = The Setsquare home page, listing its 22 diagram tools

A suite of 22 diagram and documentation tools for software work, designed and built end to end on my own. Cloud architecture builder, ER diagram builder, API flow builder, state machine designer, Venn diagrams, Gantt charts and more. Every tool opens instantly, runs in the browser, and never uploads a thing.

---
title: Supply Chain Planning Tool
domain: Product
domain: Full-Stack
badge: WORK · FORTUNE 100
badgeHighlight: true
stat: 15+ = screens built
stat: F100 = client scale
stat: E2E = design to deploy
button: Case Study -> #
note: NDA · visuals abstracted
placeholder: planning UI placeholder · 16:10\n[ abstracted screens / diagrams ]

End-to-end planning product for a Fortune 100 global beverage company, built at MathCo. 15+ interactive screens from wireframe to production.

---
title: Smart Retail Billing System
domain: Computer Vision
domain: Edge AI
badge: IEEE PUBLISHED
badgeHighlight: true
stat: 89% = detection mAP
stat: −70% = checkout time
stat: edge = fully on-device
button: Case Study -> #
link: Demo Video -> #
link: Paper -> https://doi.org/10.1109/ICOECA66273.2025.00186
placeholder: demo video placeholder

YOLOv10 on a Raspberry Pi detects products in real time and bills them through a Streamlit UI with WhatsApp notifications. Presented at ICOECA 2025.

---
title: HealthSight
domain: Computer Vision
domain: Mobile
badge: HEALTHCARE AI
stat: >90% = accuracy
stat: 3 = diagnostic models
stat: 0ms = cloud latency
button: Case Study -> #
link: Demo Video -> #
placeholder: app screens placeholder · portrait set\n[ 3 diagnostic workflows ]

On-device CNN diagnostics for foot health: flat-foot, diabetic-ulcer and fracture detection in a Java Android app with distinct workflows per category.

---
title: Multi-Crop Plant Disease Classification
domain: Computer Vision
domain: Deep Learning
badge: SCIENTIFIC REPORTS
badgeHighlight: true
stat: ViT = transformer backbone
stat: multi = crops in one framework
stat: modular = swappable heads
link: Read the paper -> https://doi.org/10.1038/s41598-026-65004-7
placeholder: leaf samples placeholder\n[ per-crop grids / confusion matrix ]

A modular framework that pairs a vision transformer backbone with classical machine learning heads, built to classify disease across several crops instead of training one model per crop. Published in Scientific Reports.

---
title: Sheep Activity Recognition
domain: Deep Learning
badge: PUBLISHED RESEARCH
badgeHighlight: true
stat: 7 = models compared
stat: 1-3s = window inputs
stat: top = LSTM-GRU + attn
link: Read the paper -> https://authors.wiley.com/index.html/article/100655703
placeholder: signal plots placeholder\n[ accelerometer windows / confusion matrix ]

Sliding-window accelerometer inputs across seven sequence models. Best result: 3-fold LSTM-GRU with self-attention, class weights and k-fold validation.

---
title: RTOS Autonomous Car
domain: Embedded
badge: EMBEDDED SYSTEMS
stat: 80% = obstacle avoidance
stat: RTOS = task scheduling
stat: bare = metal drivers
link: Presentation -> #
placeholder: hardware photo placeholder\n[ STM32 rig / wiring / run video ]

FreeRTOS prototype on STM32-F411RE Nucleo with sensor tasks, GPIO, PWM and UART drivers, and RTOS IPC mechanisms.

---
title: Automated Meme Bot
domain: GenAI
domain: Automation
badge: GENERATIVE AI
stat: 100% = hands-off
stat: 4 = services chained
link: GitHub -> https://github.com/arx-june
placeholder: pipeline diagram placeholder\n[ discovery → caption → post → QC alert ]

Fully automated pipeline: meme discovery via Google Cloud Search, Gemini caption generation, Instagram posting, and real-time Telegram quality-control alerts.