Skip to content
View henribauer's full-sized avatar
  • Berlin

Block or report henribauer

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
henribauer/README.md

Berlin skyline at dusk seen from a workbench with notebooks, charts and a laptop

Hi, I'm Henri 👋

Business student in Berlin (International Business Administration, BSP Business & Law School). I build small, practical tools for the problems I run into: studying, investing, and everyday work with documents.

Berlin Python JavaScript HTML Electron Streamlit DuckDB Obsidian

What I'm building

Project What it does Built with
how-i-build How I build software with two AI agents: the coding loop, the design track, and why each rule exists. Markdown, Claude Code, Codex
Proofloop Evidence-based audits of a website enquiry journey: every finding linked to its evidence, savings worked out from visible inputs, and each fix tested before it counts. In planning; first version due January 2027. TypeScript, Playwright, n8n (planned)
WMP website redesign Redesign prototype for the wmp-ag.de homepage: seven new sections (team, figures, services, careers, history, partner network, contact) and a template for the service pages. Work in progress. HTML, CSS, JavaScript

Also: Atlas, a macOS study companion for Obsidian, and a few private tools.

How I work

I build with two AI agents: one writes the code, the other reviews it, and I approve what ships.

  • Ideas, research, questions, then code. A new feature starts with my own ideas on paper, widened by the agent. Then agents look up principles, existing building blocks and tools, it questions me on intent and risk, and I sign off on a spec.
  • The right model for each job. The main agent plans and checks; self-contained slices go to smaller, cheaper models.
  • The builder never reviews its own work. Claude Code builds and Codex reviews, or the other way round when usage runs out.
  • Design gets more structure, not less. Style words get defined before anything is built, references come from real apps, and a prototype settles the motion.
  • Solved problems stay solved. After every tricky task the agent files a short report into a shared wiki, so the next session searches before it solves.
  • Local models on the side. I experiment with models on my own PC (RTX 4090) to see what runs offline and without usage limits.

The full workflow and the reasoning behind each rule: how-i-build.

Pinned Loading

  1. how-i-build how-i-build Public

    How I build software with two AI agents (one builds, one reviews): the coding and design workflow, and why each rule exists

  2. proofloop proofloop Public

    Evidence-based audits of a website enquiry journey: evidence → finding → change → measured result.