Boden Crouch

Summary

Self-taught engineer working on undocumented systems. I reverse engineer binary formats, build the tooling that makes them usable, and run the self-hosted infrastructure I write about. Everything I build is open source.

Selected projects

• The personal archive — Author and sole operator

  A full-provenance archive of everything I've written since 2008 — 33,184 records across 28 platforms, validated to zero schema errors.

  Python · JSON Schema · W3C-PROV · Parquet · DuckDB

• I trained an AI on 18 years of my own messages — Author and sole operator

  Can a model learn to reply the way I reply? The pipeline, the evaluation harness, and the first real numbers — including the parts where the numbers say not yet.

  Python · Local LLMs · LoRA · DPO · Hybrid retrieval

• PyKotor — Author and maintainer

  A Python library that reads and writes the file formats behind Knights of the Old Republic, so modders stop guessing at binary layouts.

  Python · Reverse engineering · Binary formats

• HoloPatcher — Author and maintainer

  A cross-platform rewrite of TSLPatcher, the mod installer the KOTOR community depends on and nobody could maintain.

  Python · PyKotor

• ModSync — Author and maintainer

  A multi-mod installer for KOTOR, for when the problem is not one mod but twenty of them in the wrong order.

  Python · PyKotor

• AgentDecompile — Author and maintainer

  An MCP server that connects AI agents to live Ghidra state, so they reverse engineer from decompilation and references instead of guessing from filenames.

  Python · Java · Ghidra · PyGhidra · MCP

• The self-hosted lab — Sole operator

  The infrastructure this site runs on, and the routing and failover work that keeps it up.

  Docker · Traefik · Cloudflare · Linux

• Synthora — Author

  A self-hosted deep research platform: multi-step retrieval and synthesis without sending the queries to somebody else's server.

  Python · LangGraph

Experience

• Software development, independent · 2011 — now

  • I take file formats and systems nobody documented, work out how they behave, and publish tooling other people can use without repeating the archaeology.
  • The long-running one is the KOTOR toolchain: a library that reads the games' binary formats, and the installers the modding community runs on top of it.
  • All of it is open source, across the [redacted], bodencrouch and bodecloud accounts.

• Agent-assisted engineering · 2024 — now

  • I work with AI assistants in the editor, and I write down what the pairing is actually good for.
  • Two guides here come out of that: the VS Code workflow, and getting a ChatGPT export out past Cloudflare.

• Digital memory infrastructure · Ongoing

  • I archived everything I have written since 2008 into one validated store: 33,184 records holding 3.49 million messages from 28 platforms, with W3C-PROV provenance on every record and zero schema errors.
  • Content-addressed storage makes re-ingestion idempotent, and Hive-partitioned Parquet keeps the full corpus queryable on my own hardware.
  • Hybrid retrieval over the top — BM25 and dense embeddings fused with reciprocal rank fusion, then reranked — so the archive answers questions instead of holding files.

• Self-hosted infrastructure · Ongoing

  • I run a Docker stack behind Traefik with Cloudflare at the edge. It serves this page.
  • Several services on it are public and free to use. Rate limits, never a paywall.
  • The supporting tools I wrote because operating it needed them are open source too.

Skills

Python · TypeScript · Docker · Traefik · Linux · Reverse engineering · Binary file formats · Next.js · Cloudflare

Education

No formal computer science qualification. The work is the evidence: bolabaden.org and the repositories linked from it.