Playing with ChatGPT RemoteMCP without OAuth
11 mins
ChatGPTs support for remote MCP servers finally makes it possible to connect custom tools and private services directly to the web interface. However, while the MCP protocol itself supports simple authentication methods, the ChatGPT web interface currently requires OAuth for non-public connectors - creating a significant barrier for experimentation, prototypes, and single-user setups. This short tutorial shows a pragmatic workaround: running a remote MCP server with a static shared secret passed via the URL, validated server-side using FastAPI middleware. While explicitly not suitable for production, this approach is ideal for learning MCP mechanics, testing tool design, and exploring integrations without standing up a full OAuth infrastructure.
Using Proof Assistants to Ground LLMs in Mathematical Truth
23 mins
Large Language Models are powerful tools for generating ideas, code, and mathematical reasoning, but they are also prone to confidently producing subtle errors. This becomes especially problematic in mathematics and theoretical physics, where correctness is non-negotiable. This article explores a practical and increasingly successful solution: combining LLMs with formal proof assistants such as Coq and Lean. In this workflow, the language model proposes definitions, lemmas, and proof strategies, while the proof assistant rigorously checks every step. The result is a system that preserves the creativity and exploratory power of LLMs while grounding every accepted claim in machine-verified mathematical truth. The article surveys recent research in LLM-assisted theorem proving, from early neural premise selection and GPT-fs contributions to Metamath, to modern agentic systems that interact step-by-step with proof environments. It then provides a hands-on tutorial showing how to guide Coq with an LLM, both manually and through automated agent loops such as Codex-based workflows. The broader outlook extends beyond pure mathematics, highlighting emerging efforts to formalize parts of theoretical physics. Together, these developments point toward a new research paradigm in which AI systems can explore bold ideas - while formal verification ensures that only rigorously proven results survive. This opens up a future where new theories can be designed and explored way more efficient and faster than before.
An Intuitive Approach behind the Lagrangian
9 mins
The Lagrangian is one of the most powerful tools in physics, yet for many of us it enters our education as something to be accepted rather than truly understood. We learn to write down L=T-V, apply the Euler–Lagrange equations, and move on - often without ever seeing why this particular combination of kinetic and potential energy is singled out in the first place. Only later, when encountering more advanced formalisms such as path integrals, does this question often resurface. This article takes a step back and reconstructs the Lagrangian again from basic Newtonian ideas. Starting from momentum, conservative forces, and the principle of virtual work, it shows how the structure and the variational statement emerge naturally without being postulated. The result is a perspective in which the action principle and the Lagrangian appears not as a mysterious axiom, but as the most compact and economical way of expressing classical dynamics and the natural bridge to its quantum generalization.
Why Use Unix Domain Sockets Instead of TCP Sockets for Backend Services Behind Reverse Proxies
13 mins
Modern backend stacks often run everything on one machine behind nginx or Apache - yet we still habitually wire internal services up with TCP ports on 127.0.0.1. This article argues that this is the wrong tool: if a service is strictly local and only meant to be reached through the reverse proxy, a TCP socket is unnecessary exposure and unnecessary complexity. Unix domain sockets are a cleaner fit for local IPC: access control becomes simple filesystem permissions, accidental exposure to the network disappears by design, and you avoid burning shared TCP resources under load. With a practical Python Uvicorn and Apache example, plus a short historical detour into why JavaEE and Tomcat ended up stuck in the everything is TCP world, this short article is a pragmatic guide for making your internal service layout safer and more explicit.
Producing Structured Output (JSON mode) with Anthropics API: A Practical Solution
8 mins
Getting structured data out of large language models is one of the key steps when building automation pipelines. While OpenAI, Mistral, and Ollama make this easy through direct response_format options, Anthropics Claude API takes a different route. There is no built-in schema parameter - yet structured, validated JSON output is still possible if one knows how to guide the model correctly. This article explains how to achieve predictable structured responses from Claude by generating a synthetic tool definition that mirrors the desired schema. The approach is elegant, compatible with other models, and turns any Claude completion into a clean, validated data exchange. A practical example demonstrates how to implement it in Python using Pydantic and how this technique bridges the gap between OpenAIs response_format and Anthropics tool-calling mechanism.
mini-apigw: A Lightweight Gateway for Multi-Model AI Infrastructure
25 mins
mini-apigw is a lightweight, locally controlled OpenAI-compatible API gateway that unifies multiple AI backends such as OpenAI, Anthropic, Ollama, vLLM, and Fooocus under one clean endpoint. It restores sanity in a world where every AI service assumes exclusive control of your GPU by adding arbitration, per-app governance, and transparent configuration through simple JSON files - no dashboards, no Kubernetes, no bloat. Designed for small labs, research environments, and hobby clusters, mini-apigw behaves like an old-school UNIX daemon: minimal, reliable, and easy to reason about. It lets you run modern LLMs and image models side by side with full control over access, security, and resources - all while keeping your infrastructure as lean and transparent as possible and using just a single API.
Ollama models not loading after update (dual RTX 3060)
5 mins
When Ollama auto-updated on my small dual RTX 3060 headless setup, model loading suddenly stopped working - no errors, just silence and hanging clients. The GPUs were detected, yet nothing generated and loading of tensors failed from one day to the other. Upgrading drivers and even moving from version 0.12.4 to 0.12.5 did not help, while other CUDA applications ran perfectly. After a few hours of debugging, the fix turned out to be simple: rolling back to Ollama 0.12.3 instantly restored normal behavior. If you are seeing lines like llama_model_load: vocab only - skipping tensors and key with type not found ... general.alignment, this post walks through what happened and how to get your models running again without tearing your hair out.
Science as a Collective, Ever-Growing Enterprise
17 mins
For a long time, I thought of science as something done by brilliant individuals - lonely minds uncovering truths in isolation, excelent researchers doing their isolated work far above the understanding capabilities of everyone else. Over time, I realized that this image is deeply incomplete. Science is not a solitary act but a vast collective enterprise: a self-organizing network of countless people across generations, each contributing a fragment to the immense mosaic of human knowledge. It’s not just about experiments, but about documenting, teaching, and preserving what we learn so others can build upon it. Yet this collective enterprise does not mean constant social interaction — its true communication channels are publications and presentations, the shared language of discovery. This essay explores science as a living, creative system - one that depends as much on imagination and diversity of minds as on logic and precision. It argues that politics and the public often miss this deeper truth: that the scientific process is society’s long-term memory and imagination combined, not merely a source of immediate economic gain or the work of some view excellent isolated minds that compete against each other and whos performance can be measured in form of economic gain or number of publications. From the quiet curiosity of basic research to the emerging cooperation between human thinkers and artificial intelligence, science remains our shared attempt to understand the universe - together.
WASM as a Secure Sandbox: From the Browser to Distributed Runners
9 mins
From faster JavaScript to a universal sandbox, WebAssembly has quietly grown into one of the most promising technologies for safe, portable code execution. This article looks at WASM not just in the browser, but as a general-purpose runtime: compiling C (or Rust) with Emscripten, running code in interpreters like wasm3, and embedding untrusted functions in distributed systems. We will also quickly look at its security strengths - and why, like any runtime, it is not a magical solution.
Building Semantic Suggested Articles for a Static Blog (and How To Visualize Embeddings)
37 mins
Semantic search is a trending topic - and as of today a suprisingly simple concept. Instead of relying on keywords, this article shows how to use embeddings to capture meaning and suggest genuinely related posts. Along the way, I will show how to visualize embeddings in 2D and 3D, making the abstract mathematics of similarity tangible through interactive plots that visualize this blog itself. From theory to practice, the article walks through a full pipeline. It covers chunking content, generating embeddings with Ollama or OpenAI, storing them in PostgreSQL with pgvector, and finally integrating related-article boxes directly into Jekyll. Code snippets, schema definitions, and configuration examples make it easy to reproduce - or adapt for your own projects.
Instrumentation Amplifiers - a quick overview
8 mins
Instrumentation amplifiers are the quiet workhorses behind precision measurement. From strain gauges in mechanical systems to Pirani gauges in vacuum technology, they allow engineers to pick up the tiniest signals while ignoring large amounts of electrical noise. Built on the foundations of the classical difference amplifier, modern INAs overcome its limitations by offering high input impedance, simple gain adjustment, and excellent common-mode rejection - all in a compact chip. This article walks through the principles step by step, starting from the difference amplifier and leading to the full instrumentation amplifier. Equations, circuit diagrams, and practical notes show why INAs are indispensable in sensor front-ends and data acquisition. Whether you’re an engineering student, hobbyist, or professional, this quick overview will give you the tools to understand when and why instrumentation amplifiers matter.
Understanding Wasps: A Misunderstood Neighbor in the Garden
58 mins
Wasps often suffer from a bad reputation as aggressive stingers, but in reality they are vital neighbors in our gardens. They act as natural pest controllers, removing flies, caterpillars, and other insects, while also visiting hundreds of plant species for nectar and contributing to pollination. Far from mindless automatons, studies show that wasps learn, remember, recognize faces, and even demonstrate reasoning abilities. Their so-called aggression is predictable and usually defensive - stings occur mainly when a nest is disturbed or a wasp feels directly threatened. By learning to read their behavior, we can safely coexist with these insects. A foraging wasp on fruit is harmless, a guard wasp with wings spread is issuing a warning, and only persistent provocation leads to an attack. Wasp colonies are temporary, vanishing in winter, but while active they provide free ecological services worth billions globally. Understanding these modes and signals helps transform fear into respect, showing wasps not as pests but as intelligent, beneficial allies in maintaining the balance of the garden.
