DEV Community: Arsen Apostolov

Building a Game for My Daughter with AI — Part 1: What If She Could Build It Too?

Arsen Apostolov — Sun, 22 Mar 2026 08:35:15 +0000

She drew a purple dragon with six legs and called it "a dragon but also a dog."

Capped the marker. Moved on. Zero doubt.

I thought: what if that drawing could actually come to life?

So I started building it. In secret.

The full story — the design thinking, the tech stack, and why the reveal moment matters more than the launch — is on Medium:

👉 https://medium.com/@arsen.apostolov/building-a-game-for-my-daughter-with-ai-part-1

Part 2 (Lovable vs Replit, MCP architecture, GPT-4 Vision pipeline) drops next week.

Follow if you want to watch a slightly obsessed AI developer build something for an audience of one.

I Hit a $200 AI Bill and Built My Own Server Instead - Complete Guide!

Arsen Apostolov — Sat, 07 Jun 2025 17:03:36 +0000

Hit a $200 Claude API bill last month ($2400 and above on yearly basis!). That was my wake-up call.

Built my own AI server instead:

RTX 3090 24GB (used): $750 one-time
Zero monthly costs
Access from anywhere via VPN
Unlimited usage

The Setup

# Install Ollama locally
curl -fsSL https://ollama.ai/install.sh | sh

# Download coding models
ollama pull qwen2.5-coder:14b
ollama pull devstral

# Use with aider
aider --model ollama_chat/devstral --api-base http://10.0.0.1:11434

Remote Access via WireGuard

The trick: secure VPN tunnel to home server.

Tech stack:

Linux server running Ollama
WireGuard VPN for encrypted access
Router port forwarding (UDP 51820)

Works from coffee shops, client offices, anywhere.

Results After 6 Months

$0 monthly bills (was $40-60/month)
Faster responses than cloud APIs
No rate limits
100% private - code never leaves my network

Want the Full Guide?

Complete walkthrough here: Stop Paying for ChatGPT - Run Your Own AI Models

Covers:

Step-by-step server setup
WireGuard VPN configuration
Router setup
Client configs for all platforms
Troubleshooting

Why This Matters

Beyond saving money, you learn:

Infrastructure management
VPN security
Cost optimization
Enterprise-ready solutions

Companies increasingly want AI that keeps data internal. This gives you both the skills and the setup.

*Connect: LinkedIn

Stop Copy-Pasting Your Code to LLMs — I Built a Tool That Does It Automatically

Arsen Apostolov — Sun, 23 Mar 2025 06:46:18 +0000

Ever wanted to ask ChatGPT or Claude about your codebase, but got tired of copy-pasting files one by one? Yeah, me too. It was driving me crazy.

Why I Built This
I was working on a project with multiple files and needed some help understanding how everything connected. So I started copy-pasting files into ChatGPT, one after another, trying to give it enough context.

It was a complete pain:

The context window would fill up
I’d forget important files
I had to manually explain connections between files
The LLM would lose track of what was what
After the third time doing this dance, I decided to just build something to solve it myself.

Find the full guide in my Medium article

Testing Aider: Practical Experience with Different Models

Arsen Apostolov — Mon, 17 Feb 2025 05:14:13 +0000

I like coding agents and all kind of supportive plug-ins to make my day easier. From Continue to Copilot. I must say that the most complete solution so far for me was Cursor, yet it is not free (see the plans).

So i found this super cool alternative - Aider chat.

I tested Aider with multiple AI backends: OpenAI, Claude, and local Ollama server.

First 30 minutes were spent learning the tool - understanding commands and workflow. After this initial setup phase, development speed increased significantly.
Model comparison from practical use:

Claude: Best performance when working remotely
Local setup: Ollama with deepseek r1 7b and qwen coder 2.5 7b
Home setup preference: Architect mode with Ollama models

Key observation: Local models provide good performance without cloud dependencies. The initial learning curve is worth the productivity gain.

What's your experience with Aider? Particularly interested in local model configurations and performance comparisons.

I’m sharing my LLM Code Lens Python package: My Secret Weapon for AI-Powered Coding

Arsen Apostolov — Fri, 10 Jan 2025 04:56:15 +0000

Hey everyone,

I want to share my tool that has completely transformed how I work with AI assistants. If you're tired of writing endless, complex prompts and struggling to get precise code insights, this is for you.

The Problem

Working with large language models like ChatGPT, Claude, or Mistral always felt like a communication battle. How do you explain an entire project's context in a single prompt?

My Solution: LLM Code Lens

I built a simple, powerful package that generates a comprehensive project context file in seconds.

Go to the output dir:

Preview the analysis.txt It contains summary of your codebase - imports, functions, documentation etc so that LLMs can have pretty good context.

Preview the full.txt It contains your codebase aggregated in a file. Great for smaller projects (bellow 10K lines of code based on my experience).

How to Use It

Install the package

pip install llm-code-lens

Generate project context

# Create condensed description of the project. Sufficient for big projects to provide context to LLMs.
# Generates analysis.txt in the output folder. Default folder: .codelens
llmcl

# On top of the analysis.txt it generates all the codebase in full.txt so that LLM can see all your projects. The output files are divided into 100K tokens file so that they can fit in any LLM including local Llama3.3 70B
llmcl --full

Get context files in .codelens/
analysis.txt
analysis.json
Paste these directly into your AI assistant

The prompt

Immediate result

Instant Efficiency!

No more:

Writing 500-word explanations
Struggling to provide context
Wasting time on prompt engineering

Just pure, efficient code insights.

My Personal Experience

This tool has been an essential part of my daily coding routine for months. It has significantly improved my development performance.

Your Turn

🚀 Star the Project on GitHub

Try it. Love it. Share it.

Hope it helps you as much as it's helped me.

Cheers,
Arsen

Let's Connect! 🤝

💼 Connect with me on LinkedIn
🎮 Join our Random42 community on Discord - AI news, Success stories, Use cases and Support for your project!
📝 Follow my tech journey on Dev.to

My Weekend project on GitHub: Making AI Art Creation Simple For Everyone 🎨

Arsen Apostolov — Sun, 05 Jan 2025 12:30:39 +0000

Check out the project on GitHub and give it a ⭐ if you find it useful!

As a developer passionate about AI, I saw an opportunity to make image generation more accessible. I wanted to create something that would enable everyone to explore the amazing possibilities of AI art without getting caught up in technical complexities. That's why I built ImageGenerator - a tool I crafted from the ground up to handle all the technical aspects behind the scenes, letting you focus purely on creativity.

Why I Built ImageGenerator Different

A key feature that sets ImageGenerator apart is that it runs completely locally on your machine. Unlike cloud-based solutions, your data never leaves your computer - there's no uploading to external servers, no privacy concerns, and no usage limits. You have full control over everything.

What also sets my implementation apart is that it's built by a developer who prioritizes both user experience and data privacy. Every feature comes from solving real problems I encountered, and I've refined the interface based on actual usage and feedback. Whether you're working on personal projects or professional tasks, you can use ImageGenerator with complete peace of mind.

Here's what you get out of the box:

🏠 100% local setup - your data never leaves your machine
🔒 Complete privacy - no cloud services or external servers needed
🎯 Simple web interface - no command line needed
🚀 One-click installation with all dependencies handled
🎨 Support for both local and online AI models
🎥 Built-in image-to-video conversion
📊 Real-time generation progress and status updates
⚡ No usage limits or API costs - generate as much as you want

See It In Action

Here are some images I generated using ImageGenerator:

Getting Started in 2 Minutes

The best part? Everything runs locally on your machine. No accounts to create, no API keys to manage, and no data privacy concerns. Just follow these simple steps:

# Clone the repository
git clone https://github.com/SikamikanikoBG/ImageGenerator
cd stable-diffusion-client

# Install dependencies
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

# Start the server
python server.py

# In a new terminal, start the client
python client.py

That's it! Visit http://localhost:7860 in your browser, and you're ready to start generating images.

Key Features That Make Life Easier

🏠 True Local Processing

All processing happens on your machine
No internet connection needed after setup
Your images and prompts stay private
Generate unlimited images without restrictions
Full control over your data and models

🎨 Intuitive Web Interface

No more juggling with command-line parameters. Everything you need is organized in clear, easy-to-understand tabs:

Connection settings
Project management
Generation parameters
Output gallery
Video conversion

🤖 Smart Model Management

Automatic scanning and loading of models
Support for both local and Hugging Face models
Easy model comparison to find what works best for you

🎥 One-Click Video Creation

Transform your still images into videos with preset animations:

Subtle movements
Normal flow
Slow motion
Ultra-slow effects

Real-World Applications

Students and developers are already using ImageGenerator for:

Creating custom artwork for projects
Generating placeholder images for websites
Experimenting with AI art styles
Building portfolios of AI-generated content

Join Our Community!

If you find ImageGenerator useful:

⭐ Star the GitHub repository
🤝 Join our Discord server for:
- Tips and tricks
- Showcase your creations
- Get help when needed
- Connect with other creators

Behind the Scenes: How I Built It

I carefully chose each technology to create the most robust and user-friendly experience possible:

Python 3.8+ for its stability and extensive AI libraries
FastAPI for a lightning-fast, modern backend that can handle heavy processing
Gradio for building an intuitive interface that anyone can use
CUDA support for GPU acceleration, which I optimized for both performance and memory usage

The entire system took several iterations to get right. I spent considerable time optimizing the model loading process, fine-tuning the memory management, and creating a seamless experience between the backend and frontend.

I'm releasing all of this under the MIT license because I believe in open source and want others to build upon what I've created. Feel free to use it in your own projects - that's exactly why I built it!

Next Steps

Ready to start creating? Head over to the GitHub repository and follow the quick start guide. Don't forget to star the repo if you find it useful!

Have questions or want to connect with other users? Join our Discord community - we'd love to see what you create!

Looking for feedback on the next article about home lab setups - AI, automation, or ML?

Arsen Apostolov — Fri, 03 Jan 2025 07:27:31 +0000

Last year, I wrote this comprehensive guide about transforming a regular home PC into a powerful learning environment using open-source tools like Linux, Anaconda, Apache Airflow, and more. The article continues to help newcomers and intermediate users get more value from their hardware.

Check it out here:

Use your (PC’s) potential — expand your knowledge | by Arsen Apostolov | Medium

Hello everyone,

medium.com

I'm planning to write Part 2, and I'd love your input on which topic would be most valuable:

Home GenAI Lab Setup: Running LLMs locally, experimenting with different models
Home Automation System: Smart home integration, custom automation solutions
Home ML Lab: Setting up for machine learning experiments and model training

What would you like to learn about? Share your preference in the comments!

Let's Connect! 🤝

💼 Connect with me on LinkedIn
🎮 Join our Random42 community on Discord - AI news, Success stories, Use cases and Support for your project!
📝 Follow my tech journey on Dev.to

5 Python Functions That Will Speed Up Your Data Analysis 🚀

Arsen Apostolov — Thu, 02 Jan 2025 05:41:17 +0000

Here are five powerful functions that can significantly boost your performance and streamline your workflows.

1. pandas.DataFrame.apply()

Transform your DataFrame operations with the mighty apply() function. By leveraging vectorization, it outperforms traditional loops when executing custom functions across DataFrame columns or rows.

import pandas as pd

# Quick demonstration
df = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]})
df['C'] = df['A'].apply(lambda x: x ** 2)

This simple yet powerful approach can dramatically reduce runtime, especially when dealing with large datasets.

2. numpy.vectorize()

Say goodbye to slow loops! numpy.vectorize() converts your Python functions into NumPy-optimized operations, perfect for efficient element-wise processing of large arrays.

import numpy as np

def my_func(x):
    return x + 10

vectorized_func = np.vectorize(my_func)
result = vectorized_func(np.array([1, 2, 3]))

Under the hood, this taps into NumPy's C-level optimizations, delivering both speed and clean code.

3. pandas.DataFrame.groupby()

Master data aggregation with groupby(). This C-optimized method distributes processing across groups, making it substantially faster than Python-level loops:

# Efficient aggregation
df.groupby('column_name').sum()

Not only does this improve performance, but it also leads to more maintainable data processing pipelines.

4. dask.dataframe

When your dataset exceeds memory limits, Dask comes to the rescue. It provides a pandas-like interface while processing data in parallel chunks:

import dask.dataframe as dd

ddf = dd.read_csv('large_dataset.csv')
result = ddf.groupby('column_name').mean().compute()

This is particularly valuable for machine learning workflows with extensive datasets.

5. numba.jit()

For computation-heavy tasks, Numba's @jit decorator is a game-changer. It compiles Python code to machine code, delivering impressive speedups:

from numba import jit

@jit
def compute_sum(arr):
    total = 0
    for i in arr:
        total += i
    return total

result = compute_sum(np.arange(1000000))

Think of it as having a C compiler at your fingertips, perfect for optimizing tight loops and complex numerical algorithms.

Wrapping Up

These functions can revolutionize your data analysis workflow. Each brings its own strengths to the table, whether you're working with large datasets, complex computations, or memory-constrained environments. Try them out and benchmark your code to see the improvements firsthand!

Let's Connect! 🤝

💼 Connect with me on LinkedIn
🎮 Join our Random42 community on Discord - AI news, Success stories, Use cases and Support for your project!
📝 Follow my tech journey on Dev.to

5 Maintenance Habits That Keep Your Dev Machine Purring

Arsen Apostolov — Sun, 29 Dec 2024 06:52:24 +0000

☕ Sunday coffee thoughts: AI is cool, but let's not forget about the basics - taking care of your hardware - your workhorse!

Just finished giving my machine some proper maintenance (pics attached) and thought I'd share some quick tips that keep these beasts running smooth:

✅ Dust is your enemy - monthly cleanup makes a massive difference in performance
✅ Clean cable management isn't just pretty - it's crucial for proper airflow
✅ Keep those drivers current - your GPU deserves that extra care
✅ Watch those temps - unexpected heat spikes are usually the first warning sign
✅ Double-check power connections - especially if you're running heavy workloads

Simple stuff, but it's what keeps our machines alive and kicking!

Let's Connect! 🤝

💼 Connect with me on LinkedIn
🎮 Join our Random42 community on Discord - AI news, Success stories, Use cases and Support for your project!
📝 Follow my tech journey on Dev.to

GitHub Copilot Now Free: why is it important? 🤔

Arsen Apostolov — Thu, 26 Dec 2024 08:46:18 +0000

The most powerful business use of generative AI - code generation at 51% - just became free for everyone with GitHub Copilot.

Recent Menlo Ventures research reveals an interesting pattern - among generative AI implementations, code generation dominates, followed by support chatbots (31%) and enterprise search (28%). GitHub's move to offer free Copilot access, including GPT-4 and Claude 3.5, signals a significant shift in the market.

⚠️ Important note: While Copilot is now free, organizations need to carefully evaluate their data privacy requirements. Free tier doesn't guarantee your code won't be used for model training - a crucial consideration for proprietary code and sensitive projects.

Looking ahead in the following years:

ℹ️ The most dominant AI use case is becoming a commodity. When your competitors have access to the same tools, success depends on integration expertise and workflow optimization.

ℹ️ Today's pattern shows businesses focusing on practical AI applications. As these become standard, the next wave will likely combine multiple AI tools - imagine Copilot-like assistance not just for code, but across your entire tech stack.

ℹ️ With current features like project-wide understanding and multi-file editing, we're already seeing AI evolve from task automation to strategic assistance. Next step? AI becoming an integral part of planning and architecture decisions.

The time for AI experimentation is over. We're entering an era where AI tools are free but expertise in using them becomes the real competitive advantage.

See the demo video I made

Let's Connect! 🤝

💼 Connect with me on LinkedIn
🎮 Join our Random42 community on Discord - AI news, Success stories, Use cases and Support for your project!
📝 Follow my tech journey on Dev.to

GitHub Copilot Now Free: why is it important? 🤔

Arsen Apostolov — Thu, 26 Dec 2024 08:36:56 +0000

The most powerful business use of generative AI - code generation at 51% - just became free for everyone with GitHub Copilot.

Looking ahead in the following years:

ℹ️ The most dominant AI use case is becoming a commodity. When your competitors have access to the same tools, success depends on integration expertise and workflow optimization.

The time for AI experimentation is over. We're entering an era where AI tools are free but expertise in using them becomes the real competitive advantage.

See the demo I made:

Let's Connect! 🤝

💼 Connect with me on LinkedIn
🎮 Join our Random42 community on Discord - AI news, Success stories, Use cases and Support for your project!
📝 Follow my tech journey on Dev.to

Running Out of Space? Move Your Ollama Models to a Different Drive 🚀

Arsen Apostolov — Wed, 25 Dec 2024 08:06:12 +0000

Is your system drive crying with 30GB+ language models? I wrote a quick guide on moving Ollama's storage location to a different drive, keeping your system partition clean.
Check out the tutorial here