Rebekka Hubert
  • About Me
  • Projects
  • Blog
Navigation bar avatar

Systems Modernization & AI Development


Thoughts on software engineering
  • Y2038: When "Impossible" Means "Wrong Approach"

    How a single compile flag replaced a failed 18-month migration attempt.

    Posted on March 5, 2026

    A 30+ year old system: dozens of statically linked 32-bit executables, tens of thousands of lines of C code, thousands of business scripts on an embedded interpreter. No tests. [Read More]
    Tags:
    • c
    • legacy-systems
    • y2038
    • systems-programming
  • Local LLMs for Code Review: Why I Don't Send Proprietary Code to the Cloud

    llama.cpp, a proprietary codebase, and the compliance department that would murder me.

    Posted on February 28, 2026

    I built an automated code review pipeline for a legacy codebase. The code is proprietary. The language is proprietary. The business logic is proprietary. Sending any of it to an external API was not an option. [Read More]
    Tags:
    • llm
    • code-review
    • local-inference
    • rag
  • Parity Testing: The Only Definition of Correct

    Unit tests can't save you. Code review can't save you. Only the database knows the truth.

    Posted on February 24, 2026

    When you translate a script from a proprietary language to Python 3, how do you know it’s correct? [Read More]
    Tags:
    • testing
    • legacy-migration
    • verification
  • Teaching an LLM a Language It Has Never Seen

    No fine-tuning, no training data, no examples on the internet. Just a spec and a prayer.

    Posted on February 20, 2026

    The proprietary scripting language in this migration has zero presence on the internet — no Stack Overflow questions, no GitHub repos, no documentation outside the company. Some claim no documentation inside the company either. It’s been running production workloads for over two decades, and the only people who know it... [Read More]
    Tags:
    • llm
    • legacy-migration
    • prompt-engineering
  • Why the Smartest AI Agent Is the Worst Translator

    How a frontier model filled my context window with brilliant thoughts and zero output

    Posted on February 17, 2026

    Migrating 6,000+ scripts from a proprietary language into Python 3. The language is old enough to vote. No LLM has seen it in training. The scripts run a legacy ERP system, and the only definition of “correct” is: the database looks identical after the translated version runs. [Read More]
    Tags:
    • llm
    • legacy-migration
    • ai-tooling
  • ← Newer Posts
  • Email me
  • GitHub

Rebekka Hubert  •  2026

Powered by Beautiful Jekyll