Experiments
Just some experiments I conduct when reading articles or when stumbling across interesting ideas.
Adding German Court Judgments to My Public-Data RAG System
A follow-up to my public-data RAG kickoff: why I added German court judgments as a new source type, why I replaced LLM chunking with a deterministic pipeline for structured documents, and what broke once the corpus reached real scale.
Running Hermes Agent on IONOS with OpenRouter and Kimi 2.6
A short practical setup note for running Hermes Agent on a low-cost IONOS VPS with SSH access and OpenRouter budgeting.
Building a Public-Data RAG System End-to-End
An engineering case study of an end-to-end public-data RAG system: using legal research as the example workflow to showcase ingestion, retrieval, evaluation, and the next iteration path.
Building Lisa, an Insurance Claims Voice Agent, at Big Berlin Hack
How we built Lisa, a phone-based insurance claims voice agent, during Big Berlin Hack: what made the demo convincing, what was hard about live audio, and what I learned about building voice agents quickly.
Getting Books into My AI/ML Knowledge Base
A practical workflow note on turning books into chapter-sized inputs, summarizing them incrementally, and folding the result into my AI/ML wiki without burning too many tokens.
Using My Knowledge Base to Build a Harness Engineering Podcast
A practical workflow note on using my AI/ML knowledge base to turn podcast references, related sources, and wiki context into a new 60-minute NotebookLM podcast on harness engineering.
Building an LLM-Maintained AI/ML Knowledge Base
A case study of an Obsidian-based AI/ML knowledge base where raw sources, generated summaries, and a concept wiki are maintained incrementally with Claude Code.
How I Turned a Transcript Workflow Gap Into an API Product
How I identified a gap in agent-first podcast transcript access and turned it into Podfetcher: an API product with billing, docs, SDK, CLI, and MCP integration.
Getting Email Access for OpenClaw: Dead Ends, Agent Mail, and a Custom CLI Wrapper
How I added email access to my OpenClaw workflow by testing provider options, discovering IMAP limits in Agent Mail, and shipping a custom CLI wrapper around the Agent Mail SDK.
My OpenClaw Journey (So Far): Starting Late, Choosing a VPS, and Switching to AWS
A practical write-up of setting up OpenClaw on a low-cost AWS EC2 instance, fixing memory issues, and shaping it into an AI morning briefing workflow.
Building a Tiny Coding Agent – Plan, Edit, Verify, Repeat
A small local coding agent that safely edits files using a simple plan–act–observe loop, diff previews, and optional test verification.