news

Dec 25, 2025 I’ve published a new article, “GPT-5.1 vs Ministral-3 3B: Evaluating AI Newsletter Quality For Local AI Newsletter Generation”, where I test whether a $0 local model can match a frontier cloud model for automated newsletters. I ran the same 6-stage n8n workflow—news ingestion, relevance scoring, editorial decision, research enrichment, formatting, and delivery—with both GPT-5.1 and Ministral-3 3B (via Ollama on a MacBook). Using LLM-as-a-judge evaluation across five topics, the local model scored 3.47 vs 4.07—closer than expected. The post covers three techniques that make local models viable: task decomposition, JSON auto-correction, and short context windows. You can read it on Akribic here.
Nov 26, 2025 I’ve published a new article, “AI News Agent: A Smarter Newsletter That Knows When to Send”, where I share an intelligent n8n workflow that decides when news is actually worth sending. Unlike fixed-schedule newsletters, this AI News Agent uses an editorial AI to evaluate whether today’s news meets quality thresholds—with built-in deduplication against the last 5 editions and configurable frequency guardrails (min/max days between sends). The workflow fetches news via SerpAPI’s DuckDuckGo News, stores articles with automatic deduplication, and enriches selected pieces with Tavily web search before delivering via Telegram. You can read it on Akribic here.
Nov 20, 2025 I’ve published a new article, “How I built my own fully-automated weekly news digest - here is how you can do that too.”, where I walk through how to build an AI-powered newsletter workflow in n8n. The workflow pulls fresh articles via NewsAPI, uses OpenAI GPT-5.1 to select the 5 most relevant ones, enriches them with Tavily Search, and sends a clean, compact newsletter straight to Telegram—essentially acting as a research assistant on autopilot. The post includes a downloadable workflow (ai-newsletter-n8n) and step-by-step instructions so you can adapt it to your own topics and setup. You can read it on Akribic here, or on Medium here.
Nov 20, 2025 I’ve launched Akribic, a new blog focused on hands-on AI automation workflows. On Akribic, I share step-by-step breakdowns of real automations, concrete configs and snippets you can reuse, and honest notes on what actually works (and what doesn’t) when building AI-powered workflows in practice. You can check it out here: https://www.akribic.com.
Nov 27, 2024 SafePlates is now available on safeplates.de! SafePlates is a recipe assistant that helps create personalized, allergen-safe meals. It lets you create personalized recipes based on your dietary needs. It now features
  • A visual depiction of the meal
  • Refined recipe generation
  • Automatically extracted metadata from the recipe such as prep time, number of servings, and nutritional information.
Check it out on safeplates.de! 💻
Oct 28, 2024 I’m happy to share the launch of SafePlates, an AI-powered recipe assistant that helps create personalized, allergen-safe meals. 🌱🍽️ Built with OpenAI GPT-4 and using Langchain/LangGraph, SafePlates checks for allergens and suggests safe alternatives. It also features an easy-to-use Gradio interface and is now available on Hugging Face Spaces. Check it out on Hugging Face! It’s free and open-source! 💻
Oct 17, 2024 I’ve just published a blog post exploring how LlamaParse and Multimodal LLMs allow us to extract insights from complex PDFs containing both text and images. 📄✨
In the post, I walk through:
  • 🔍 Parsing documents with both text and images
  • 🤖 Using GPT-4V to interpret and query the parsed content
  • 📊 An example: analyzing U.S. election results from 2016 and 2020.
Intelligent PDF parsing + multimodal LLMs are really mighty for document processing, allowing us to handle even more than just plain text—like images, charts, equations, … 🚀
See the full post here! 💻
Mar 19, 2024 We just published a new paper on how to properly initialize reservoir computers when only a limited history of observations is available. Check out the code here and the paper here.