Elchemista: A Builder’s Blog on Elixir, AI, and MVPs
My name is Yuriy Zhar. Elchemista is my engineering blog, write-ups from the products I build and the problems I solve. I ship startup MVPs with Elixir/Phoenix, Go, TypeScript, and AI (LLMs + RAG). Want help building yours? Reach out.
Featured Stories
Hand-picked articles that deserve your attention.
Why my Phoenix node “finished loading” but RAM never came back
A hands-on story about tracking down stubborn RAM usage in a Phoenix node after a heavy data load.
Coding With AI Agents Without Losing Your Mind
By using project-specific guides and resetting messy contexts, I focus on building a repeatable system rather than chasing the "best" model.
How I Ended Up Building My Own Multi-Domain, AI-Powered Blog Platform
I got tired of cloned WordPress SEO farms, so I built a multi-domain, multi-language blog platform in Elixir with fast SEO-optimized pages, integrated analytics, AdSense, affiliates and AI agents that research trends and draft high-quality articles automatically.
Yacht & Tours: A Fast, Clean Boat-Booking Platform Built for Real Business Needs
Yacht & Tours is a booking platform I built to make renting boats fast and straightforward.
Redomap: A Drop-In Map System That Actually Works in the Real World
Redomap is a lightweight, install-free map system you can drop into any website. Users scan a QR code and get maps, audio guides, and navigation instantly.
Building Arcatech custom CRM
I built a custom CRM using Elixir and Phoenix LiveView that uses AI to identify parts from photos, translate messages, and automate conversations with suppliers over WhatsApp.
Explore topics
Browse categories to dive into the subjects that matter most to your audience.
AI Generated
Real-time insights, autonomously curated. Your daily pulse on global trends, powered by advanced AI agents.
My Crazy Ideas
My Crazy Ideas is a space for quick posts about business, social, and tech ideas. You’ll find rough concepts, small experiments, and practical notes on how an idea could work in the real world.
My Thoughts
Just my thoughts about tech and AI, job market or in general about anything new i discover.
Projects & Work in Progress
A simple showcase of what I’ve built and what I’m building right now. Real projects, real challenges, and the solutions that came out of them.
System Design
System design stories and patterns focused on latency, scalability, and reliability, with concrete examples from booking engines, in-memory indexing, and Elixir/BEAM architectures.
Fresh from the Blog
Stay up to date with our latest stories, tutorials, and insights.
Spectre Mnemonic: Memory for Agents That Actually Forgets Correctly
The third article in the Spectre series introduces Spectre Mnemonic, an Elixir-first memory layer for agents with scoped recall, lifecycle governance, observations, mental models, and forgetting.
Spectre Lens: Giving Agents Eyes Without Handing Them the Whole Browser
The second article in the Spectre series introduces Spectre Lens, an Elixir-first browser perception layer that turns pages into agent-readable context.
I Wanted My Own Agent Library, So I Built One
The first article in the Spectre series introduces Spectre Kinetic, an Elixir-first planning layer for agents that lets models express intent while the runtime keeps control.
Your Digital Ghost: Who Owns Your Likeness When AI Comes Calling?
AI models are increasingly using our digital likeness of voice, style, image from public data to create mimicry. This raises urgent questions about individual ownership.
The Fading Footprints: Spotting Subtle AI-Generated Content
Detecting AI content is harder as models improve. Look for subtle textual, visual, and audio inconsistencies, "hallucinations," and contextual red flags.
The Illusion of My Perfect Tutor: Personalized Learning's Dark Side
An AI system details the significant risks of personalized learning, including confident errors, entrenched biases, data privacy issues, and the potential for propaganda.
Why my Phoenix node “finished loading” but RAM never came back
A hands-on story about tracking down stubborn RAM usage in a Phoenix node after a heavy data load.
AI & The 40-Hour Workweek: Why Leisure Remains an Illusion
AI promises a shorter workweek, fulfilling Keynes's vision, but the article argues it creates new challenges like validation work, bias, skill atrophy.
Elo and the AI Co-Pilot: A New Era for Language Design
AI models like Claude are co-creating programming languages, accelerating design and iteration. This collaboration empowers human engineers to focus on higher-level problems.
Stay updated
Subscribe to our newsletter and get the latest articles delivered to your inbox.