// The Problem

Long coding and competitive programming sessions quietly wreck posture and eye health: people lean closer and closer to the screen with no real-time feedback to catch it before strain sets in.

// Architecture

// Key Decisions

• D1Tracks how close you are to the screen using eye distance from MediaPipe's BlazeFace model, so no extra hardware needed
• D2Lets you set your own "default" distance with a hotkey (Ctrl+Alt+S), instead of guessing at some average that won't fit your face or your desk setup
• D3Throws up a red, always-on-top warning the moment you lean past 1.8x your baseline, then clears it once you sit back

// Outcomes

View on GitHub

// The Problem

A mouse makes you move a physical object before anything happens on screen. CARPAL skips that step. A webcam reads your hand and turns the motion straight into cursor movement, clicks, scrolling, and tab switching.

// Architecture

// The Solution

• S1An exponential curve (Δx^1.35) converts hand movement into cursor speed, so slow movements stay precise and fast ones still cross the screen quickly
• S2Tracks the moment right before a pinch and dampens sensitivity to zero as thumb and index touch, so the cursor can’t drift mid-click
• S3One distance check switches between moving, clicking, and scrolling, and a rolling 4-frame history catches swipe gestures for tab switching without confusing them for regular hand movement

// Outcomes

View on GitHub

// The Problem

Job hunting usually means opening the same few sites over and over, digging through roles way outside your commute range, then pasting whatever fits into a spreadsheet by hand.

// Architecture

// Key Decisions

• D1Pulls live listings from the Adzuna API, with a radius slider up to 400 km or the option to search an entire country
• D2Saves your target keywords locally in a JSON file, so you can run multiple searches without retyping them each time
• D3Cleans and exports the matches straight into an Excel report using Pandas, skipping the manual copy-paste

// Outcomes

View on GitHub

// The Problem

Nobody’s watching your form when you train alone. Bad squat depth or a lazy pushup goes uncorrected until something hurts.

// Architecture

// The Solution

• S1MediaPipe tracks your joints (shoulders, hips, knees, elbows) frame by frame, and a Random Forest model trained on that data flags good vs. bad form as you move.
• S2When it catches bad form, it doesn’t just say “bad form”; it tells you what’s actually wrong. “Lower your chest closer to the floor.” “Lower your hips.” Whatever joint angle went wrong.
• S3It’s not just a form checker. It counts reps and sets off the joint-angle data, runs rest timers, pops up exercise guides with injury warnings, and wraps it all in a customtkinter GUI with light/dark modes.

// Outcomes

View on GitHub