Skip to content

Blog

Notes on software development, SvelteKit, Directus, automation and AI.

15 min read

The NK model, step by step, with nkscape

What an NK fitness landscape is, explained with a firm that has three decisions: payoff tables, peaks, why step-by-step improvement gets stuck and how it all looks in nkscape, my Python library.

  • NK model
  • Agent-based modeling
  • Python

4 min read

Using Proxyman to reverse engineer the Instagram private API

The Instagram API traffic is encrypted using SSL/HTTPS. In this post I explain how to build a modified Instagram APK and use Proxyman and a custom certificate to intercept the traffic sent by the app.

  • Proxyman
  • Reverse Engineering
  • Instagram API

3 min read

Lazy-Loading Images and Components in Svelte and SvelteKit with Typescript

Lazy loading components and images in SvelteKit is really easy. It can be achieved by simply defining a custom action modifier. In this post I walk through every step of the process!

  • Svelte
  • Typescript
  • Lazy Loading

Written for S&S Technologies GmbH

Posts I wrote for our company blog about n8n, AI agents and language models. They open on sus-tech.com.

  1. opens on sus-tech.com

    Matrix42 MCP Server: Ask Your Service Desk in Plain Language

    An open-source MCP server that lets Claude, Copilot or Cursor work with Matrix42: the API, the data model and the service desk. Read-only by default.

  2. opens on sus-tech.com

    GPT-OSS: OpenAI's Open-Source Models That Rival Proprietary LLMs

    OpenAI's gpt-oss-120b and gpt-oss-20b are open-weight models you can run yourself. What they mean for AI agents that work without a cloud API.

  3. opens on sus-tech.com

    Machine Learning vs. Generative AI: What SME Leaders Need to Know

    Machine learning predicts, generative AI creates. The difference in plain terms and which business tasks suit each in a small or mid-sized company.

  4. opens on sus-tech.com

    What Is Artificial Intelligence? A Plain-English Guide for SME Leaders

    A plain-English introduction to artificial intelligence for SME leaders: what it is, where it comes from and where it pays off in daily business.

  5. opens on sus-tech.com

    Prompt Engineering Is Not Enough

    Clever prompts make good demos, not reliable processes. What AI in production needs besides: retrieval, fine-tuning and workflows around the model.

  6. opens on sus-tech.com

    LLM Finetuning: LoRA and QLoRA

    How LoRA and QLoRA fine-tune large language models with a fraction of the GPU memory and what that means for the cost of custom models.

  7. opens on sus-tech.com

    RAG vs. Finetuning: Choosing the Right Strategy for your LLMs

    Retrieval-augmented generation or fine-tuning? When to let a language model look up your data and when to train it, with examples from IT and ERP.

  8. opens on sus-tech.com

    n8n vs Make: Choosing the Right Automation Engine

    n8n or Make? A comparison of pricing, GDPR and AI options and why self-hosted n8n with local models can cut costs and latency.

  9. opens on sus-tech.com

    n8n vs. Zapier: Choosing the Right Automation Engine

    n8n or Zapier? A comparison of pricing, GDPR, AI costs and hosting and what changes when n8n runs on your own server with Ollama.

  10. opens on sus-tech.com

    Matrix42 Integration in n8n: S&S Technologies' n8n-nodes-matrix42

    n8n-nodes-matrix42 connects Matrix42 to n8n: installation, credentials, the operations of the node and an end-to-end example.

  11. opens on sus-tech.com

    Cloud LLM vs Local AI-Agent: Cost Comparison with Real Example

    What an invoice-processing agent costs with a cloud LLM API and with a small model on your own GPU, calculated on a real workflow.

  12. opens on sus-tech.com

    The Cost of Doing Nothing: Hidden Inefficiencies in Legacy Workflows

    Copying data between ERP and spreadsheets by hand costs more than it seems. Where the hidden costs are and how automation with n8n removes them.

  13. opens on sus-tech.com

    Small Language Models (SLMs): Agentic AI for Your Workflow Automation

    Most AI agent tasks run well on models under 10 billion parameters. Why small language models on Ollama and n8n cut costs and keep data in-house.

  14. opens on sus-tech.com

    Internal Knowledge Bases with RAG: Architecture, Tools and Maintenance

    How to turn PDFs and documents into a chat that answers with sources: ingestion, embeddings, vector store and upkeep of a RAG knowledge base.

  15. opens on sus-tech.com

    GDPR & AI: Safe Local Hosting for LLMs

    Why cloud LLMs are a GDPR risk for European companies and how running language models locally keeps data, costs and latency under control.

  16. opens on sus-tech.com

    How to Run LLMs Locally

    Step by step: choose a model, check your hardware, install Ollama and connect it to n8n to run language models on your own machines.