There's never been a better time to build. The world now runs on natural language — and I speak it fluently. I don't memorize syntax. I architect ideas. I don't just use AI — I guide it, with intention and intuition.
Every developer's biggest challenge? LeetCode problems.
The real challenge isn't the code—it's the lack of true understanding. Most resources don't provide the deep knowledge needed for that magical "aha!" moment.
I'm solving this first for myself—like the airplane safety rule: put on your own oxygen mask before helping others. Once I truly understand, I can help others see the light too.
Inspired by @naval's wisdom: "It's not 10,000 hours, but 10,000 iterations."
But do we really need 10,000? Maybe it's 108 or 1,008—like the sacred repetitions in Hindu mantras. Ancient wisdom suggests these numbers permanently encode knowledge into our consciousness.
I built this video segment looper for myself—to master concepts through deliberate repetition. While I can't release it publicly due to content rights, if you're interested in the concept, reach out. Together we could create a platform where people upload their own content, YouTube-style, without copyright concerns.
📋 Copy links to share:
Want more details? Ask ChatGPT about me. It knows a few things 😉
Fresh tricks and hacks that are saving devs hours every week.
The workflow that 10x'd my productivity
Get into flow state in 30 seconds
Fresh drops from the workshop. Still warm from the code oven!
50+ new automation scripts
My complete development environment
I didn't start with a grand plan—I started with a few scrappy scripts. A commit message suggester here, a README fixer there. Each assistant helped remove friction from my workflow, and slowly, they formed an AI loop that felt... alive.
These weren’t just ChatGPTs. They were finely tuned bots that could triage issues, summarize bug tickets, or even remind me what I was thinking last Friday at 2 AM.
Eventually, this ecosystem needed a way to practice speech, review audio snippets, and remember more of what I cared about. That led me to create Segment Loop Master — a project born entirely from this GPT-first workflow.
People often say it takes 10,000 hours to master something. But the modern mystic @naval reframed it: it’s not about hours—it’s about 10,000 iterations.
I wanted to count and refine those iterations. I imagined a platform where each 20-second clip from a podcast or video could become a micro-lesson—chewable, masterable, replayable. If I gained insight, I ’d record it. If not, I’d move on.
That led me to ask Mimi to build this foundational script:
#!/bin/bash
if [ -z "$1" ]; then
echo "❌ Usage: $0 <filename.mp3>"
exit 1
fi
INPUT="$1"
BASENAME="${INPUT%.*}"
SEGMENT_DURATION=20 # in seconds
# Get total duration in seconds
DURATION=$(ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "$INPUT")
DURATION=${DURATION%.*}
START=0
while [ "$START" -lt "$DURATION" ]; do
END=$((START + SEGMENT_DURATION))
if [ "$END" -gt "$DURATION" ]; then
END=$DURATION
fi
MINUTE=$((START / 60))
SECOND_IN_MINUTE=$((START % 60))
NEXT_SECOND_IN_MINUTE=$((SECOND_IN_MINUTE + SEGMENT_DURATION))
if [ "$NEXT_SECOND_IN_MINUTE" -gt 60 ]; then
NEXT_SECOND_IN_MINUTE=60
fi
OUTPUT="${BASENAME}-m${MINUTE}-${SECOND_IN_MINUTE}s-${NEXT_SECOND_IN_MINUTE}s.mp3"
ffmpeg -hide_banner -loglevel error -ss "$START" -t "$SEGMENT_DURATION" -i "$INPUT" -acodec copy "$OUTPUT"
echo "✅ Created $OUTPUT"
START=$END
doneThis was the spark behind what became Segment Loop Master.
My collection of specialized AI assistants, each trained for specific tasks.
Ruthless but fair PR reviews
Makes README files actually readable
Creative ways to integrate GPT APIs into workflows that actually matter.
Never write "fixed stuff" again
From crash to ticket in seconds
The bots that work while I sleep. My passive income generators.
Posts my projects automatically
Sorts GitHub issues like a pro
Claude isn't just an AI assistant to me — it's my coding buddy, rubber duck, and sometimes the voice of reason when I'm about to commit something questionable at 2 AM.
The secret sauce prompts that turn Claude into a coding wizard. Spoiler: It's all about context and personality!
My daily workflow with Claude. From planning to debugging, we're like the Batman and Robin of development.
Advanced techniques that make other devs go "Wait, how did you do that?"
Sam:
"Claude, help me refactor this React component to be more performant"
Claude:
"I see some optimization opportunities! Let's start with memoization..."
Inspired by @naval's wisdom: "It's not 10,000 hours to master a concept, but maybe 10,000 iterations." This sparked the idea for Segment Loop Master - a tool to play video segments repeatedly while documenting insights with iteration numbers.
Kept it deliberately simple for easy querying with MongoDB Compass. Used integer source_id for debugging instead of ObjectIds.
// Segment document structure
{
source_id: 1, // Integer for easy debugging
segment_name: "intro-0-20s",
filepath: "/path/to/segment.mp4",
start_time: 0,
end_time: 20,
iteration_count: 0,
insights: []
}Created a zsh script to split videos into 20-second segments for focused learning.
# Split video into 20s segments
$HOME/zScripts/bin/music/split_video_by_minute_segment.sh
# Creates segments like:
# video-m0-0s-20s.mp4
# video-m0-20s-40s.mp4
# video-m0-40s-60s.mp4Asked Windsurf to create a Python script to load segment data by providing the video path. It successfully generated the data loader.
Requested a Python FastAPI backend with React frontend. After hours of attempts, Windsurf couldn't create a functional video player. Too many moving parts.
Ditched React to reduce complexity. With vanilla JavaScript, finally achieved a basic functioning player. Simplicity won.
"Context engineering is more important than prompt engineering"
Every time I tried to fix something, Windsurf would run into CORS issues - even though CORS was fixed in the early stages. The agent had lost context of previous solutions.
This was my deep realization: For agent-based development tools, maintaining context is paramount. Without proper context, agents repeat mistakes, forget solutions, and spiral into circular debugging patterns.
The quality of your context determines the quality of your AI-assisted development.
The timeline progress bar wasn't responding. Hours of debugging led to a crucial insight: split functionality into separate components so fixes won't cascade into other parts.
// Component separation strategy
- mediaPlayer.js // Video player logic
- segmentList.js // Segment listing
- timeSkipper.js // Timeline controls
- sourceManager.js // Source managementAdded web interface for path loading, but it corrupted my data model by creating ObjectId references. Realized I needed to give extremely detailed instructions to Windsurf.
Solution: Created README-encountered-issues.md to track all fixes
Realized that requests often go to new server instances with no knowledge of previous fixes. Created comprehensive README files as context anchors:
Using these README files, tested creating new backends with Node.js and Java. The development was remarkably quick - context made all the difference.
With proper Context Engineering + Precise Prompt Engineering
The game-changer for debugging: Frontend sends sequence logs to backend with incrementing numbers.
// Frontend sequence logging
frontendLogger.log('Segment selected', {
segment_id: segment._id,
name: segment.name
});
// Outputs:
// [FRONTEND DELUX:1] Segment selected
// [FRONTEND DELUX:2] Loading media...
// [FRONTEND DELUX:3] Playback started# Backend receives and logs with sequence
@app.post("/api/log_sequence")
async def log_sequence(data: dict):
message = data.get("message")
delux_backend_logger.log_message(message)
# Outputs:
# [0] Logger initialized
# [1] [FRONTEND DELUX:1] Segment selected
# [2] [FRONTEND DELUX:2] Loading media...Screen recording showing the Segment Loop Master interface with media player, segment controls, and iteration tracking
Behind the scenes look at the Segment Loop Master source code implementation
Document everything. Your future self (and AI) will thank you.
Separate concerns to prevent cascade failures.
Unified frontend/backend logging reveals the truth.
ZScripts is my collection of automation magic. These aren't your basic shell scripts — they're productivity powerhouses that handle the boring stuff so I can focus on the fun stuff.
Keyboard shortcuts that do impossibly complex things. It's like having superpowers.
The scripts that run my entire development workflow. Set it and forget it!
The secret weapons that make me look like I have 48 hours in a day.
The repositories that actually solve real problems. No todo apps here!
Control VLC with natural language. "Play that funny scene from The Office" actually works.
React dashboard that actually makes IoT devices useful instead of annoying.
The fun stuff I build when I should be sleeping. Some became actual businesses!
Tracks your coding mood and suggests the perfect playlist. Spotify integration included.
Chrome extension that gives you creative excuses to leave boring meetings.
After 20+ years of tweaking, optimizing, and probably over-engineering, I've finally achieved developer enlightenment. Here's the setup that keeps me in flow state.
The missing package manager for macOS. Install and manage all CLI and GUI tools.
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"Manage multiple runtime versions (Java 23, Python 3.9.7, Node 23.2.0)
brew install asdf
asdf plugin add java
asdf install java adoptopenjdk-23.0.1+11The terminal emulator that makes Terminal.app look like a toy
Rainbow-style prompt with git-aware features
git, zsh-autosuggestions, zsh-syntax-highlighting
# Quick navigation
alias cdpmr="cd /Users/sammuthu/Projects/MetabolicReset"
alias cdscripts="cd $HOME/zScripts/bin"
# Git shortcuts
alias gs="git status"
alias egcp="$HOME/zScripts/bin/git-commit-push.sh"
# System utilities
alias refz="source ~/.zshrc"
alias versions="asdf current"
alias myip="ipconfig getifaddr en0"With Claude Code terminal integration
Codeium's AI pair programming
Documentation & insights
Coding & automation
ollama run llama2Local LLM GUI for experimentation
Custom macros for everything:
Spotlight on steroids:
Local instance for all hobby and POC projects
# My MongoDB helper scripts
$HOME/zScripts/bin/mongo-backup.sh
$HOME/zScripts/bin/mongo-restore.sh
$HOME/zScripts/bin/mongo-clone.shGUI for managing and visualizing collections
Yes, I still use Vim. OneDark theme with modern UX settings.
" OneDark theme with vim-plug
call plug#begin('~/.vim/plugged')
Plug 'joshdick/onedark.vim'
call plug#end()
" Modern settings
set number relativenumber
set smartindent
syntax on
set clipboard=unnamed " System clipboard
" Custom shortcuts
nnoremap <Space> :
nnoremap <C-s> :w<CR>My dotfiles are symlinked from ~/zScripts/bin/dotfilesto keep everything version controlled and synced.
# Sync all dotfiles
$HOME/zScripts/bin/sync_dotfiles.shOver the years, I've built various tools and utilities. Some are decades old but still incredibly useful!