nullC C Compiler in C Lexer: 100% Parser: 30% nuLLM LLM from Scratch Architecture: 100% Chat ready: 90% JoshLang Custom Language Design: 100% Implementation: 0% Word Bot Etymology Tool 4 commands 100% done Automation & Infrastructure • Morning brief: Weather + astrology • Finn auto-update script • GitHub trending monitor Documentation & Publishing • Journal: heyitsmejosh.com/journal/ • Portfolio: Added blog link • All READMEs + map.svg • GitHub repos: 4 new + updated Tools & Skills Added project-mapper • word-bot • lobster-anatomy Slash commands: /define, /etymology, /synonym, /rhyme Session Stats: Token usage: 83.6% (167k/200k) • Projects: 4 created • Commits: 40+ • Lines of code: ~2000+ Time: ~4 hours • Output: Fully documented, tested, and deployed

Started this journal. Should probably write down what I build instead of letting it vanish into commit logs.

nullC - C Compiler in C

Building a compiler from scratch. Not because I need one - because I want to actually understand how they work.

Lexer’s done (tokenizes C source). Parser’s 30% there (handles simple programs - hello world, arithmetic, variables). 10 complexity levels mapped out, ending with a meta-compiler (compiler that compiles itself).

Goal: compile a TCP server, then self-host.

github.com/nulljosh/nullC

JoshLang - Custom Programming Language

Started designing my own programming language. Not C, not Python - JoshLang.

Features: no semicolons, string interpolation ($var), pattern matching, optional types, implicit returns. Clean syntax that reads like thoughts. 5-stage plan: design, lexer, parser, interpreter, compiler.

Design complete. Implementation next.

github.com/nulljosh/JoshLang

Word Bot - Etymology & Definitions

Built a word lookup tool with four commands: /define, /etymology, /synonym, /rhyme. Uses Dictionary API and Datamuse for real-time lookups.

Works in chat - type /define universe and get definitions with examples. Rhyme command returns actual rhyming words (universe → terse, diverse, inverse…).

Useful for writing, poetry, or just curiosity.

nuLLM - Minimal Language Model

Also started building an LLM from scratch. Same philosophy - understand transformers by building them.

Complete architecture implemented: tokenizers (char + word), attention (single + multi-head), full transformer blocks, training loop, generation with temperature sampling. Comprehensive test suite passing. 9/10 complexity levels done.

All that’s left: install PyTorch and it’ll actually generate text.

github.com/nulljosh/nuLLM

Other Stuff

  • WikiScroll: Added random button. Sometimes you don’t want to scroll linearly.
  • GitHub trending: Daily digest of what blew up.
  • Finn auto-update: Script to update finance tracker without manual HTML edits.
  • Morning brief v2: Weather + trending + tasks + calendar in one check.

Small automation wins. Less human spreadsheet, more actual work.

Next

  • nullC to level 5 (function calls)
  • Keep shipping

This journal exists because I forget what I worked on yesterday. Memory files help Claude. This helps me.