Writing
Essays, notes, and code.
I write to think out loud. Mostly about LLMs, agents, and whatever I'm currently breaking.
Apr 23, 2025
Agent2Agent and MCP: An End-to-End Tutorial for a complete Agentic Pipeline
This tutorial will guide you through the process of building a complete agentic pipeline using Agent2Agent and MCP. We will cover the entire pipeline, from data preparation to model training and evaluation.
→
Jan 24, 2025
Free AI, Biased Answers: A Deep Dive into China’s Deepseek and the Perils of Propaganda
A critical look at how Deepseek, a new Chinese-developed LLM, selectively censors and manipulates language, using propaganda tactics and silencing uncomfortable truths, all under the guise of offering free AI services
→
Nov 13, 2024
ErisForge: Customizing LLM Behaviors for Enhanced Control and Research
ErisForge empowers developers to adjust refusal, tone, and other behaviors within LLMs, offering a versatile toolkit for customization, adversarial testing, and research on model censorship.
→
Feb 27, 2024
When LLMs confess: Prompt Injection and Data Exfiltration
Unveiling the risks and defenses against prompt/data exfiltration attacks targeting Large Language Models (LLMs), this comprehensive exploration sheds light on how attackers can manipulate LLMs to divulge sensitive information and outlines robust strategies for safeguarding these AI systems
→
Dec 15, 2023
Enhancing Your Haystack Experience: The Unofficial Chatbot Guide
The Haystack Documentation Chatbot is an LLM agent that you can use to chat with the documentation of the famous NLP framework Haystack.
→
Nov 12, 2023
Kompy - A Python Wrapper for Komoot APIs
Kompy is a Python wrapper for Komoot APIs, designed for outdoor enthusiasts and developers who want to access and manipulate their Komoot data for personal dashboards or analysis. Easy to install and use, it allows for downloading and uploading activities from and to Komoot, making it a versatile tool for integrating outdoor activities with digital platforms.
→
Oct 26, 2023
Step Back Propting - Let's make LLM think about questions
Exploring the transformative potential of STEP-BACK PROMPTING, the paper introduces a novel, human cognition-inspired technique aimed at substantially enhancing the multi-step reasoning abilities of Transformer-based Large Language Models, showcasing significant performance improvements across varied complex tasks.
→