Jerome Hardaway.

Artifacts

Everything here is work I did at Vets Who Code. Each entry states an outcome, and each deep dive covers the problem, the architecture, the tradeoffs, the evidence and what I would change in a second version.

  • 5.1 to 1.2points of AI grading spread, across three eval rounds

    Hashflag Skills

    Cut AI résumé-grading spread from 5.1 points to 1.2 with a three-round eval.

    Makes an AI-generated score stable enough to give a job seeker advice from it.

    • LLM evaluation
    • Claude Code
    • Python
    • TypeScript
    • CI on every push
    • eval run October 2026
    • 98–100% grader agreement

    Deep diveRepoEval

  • 388 of 389commits on a production AI backend are mine

    J0DI3 AI backend

    Built a production AI teaching backend alone, with 670+ tests.

    Gives every veteran in the program coaching on demand without giving away the answer.

    • Python
    • FastAPI
    • RAG
    • PostgreSQL
    • pgvector
    • Terraform
    • 70% coverage floor in CI
    • runs on Cloud Run
    • token budget per learner

    Deep diveCode walkthrough on request

  • $0in API keys to run a 7-skill, 7-agent job search kit

    AI-Era Job Kit

    Shipped a 7-skill, 7-agent job search kit that runs on $0 of API keys.

    Lets any job seeker score their résumé against live labor data and rerun every number for free.

    • AI agents
    • Claude Code
    • Python
    • Public data APIs
    • 62 tests passing
    • v1.1.0 released November 2026
    • no dependencies

    Deep diveRepoWorkshop

  • 4,455military job codes translated into tech careers

    MOS Translator

    Turned 4,455 military job codes into tech career guides with zero model cost per view.

    Answers the first question every transitioning veteran asks: what does my job code qualify me for?

    • Data pipelines
    • LLM pipelines
    • Gemini API
    • TypeScript
    • Structured output
    • schema-validated output
    • shipped May 2026
    • static pages

    Deep diveLive demoSource

  • 31 MBserver bundle, down from a deploy that failed at 250 MB

    vetswhocode.io platform

    Took a failing deploy from over 250 MB to a 31 MB server bundle and kept 215 routes shipping.

    Runs the front door of a nonprofit that has trained more than 300 veterans.

    • Next.js
    • TypeScript
    • React
    • CI/CD
    • Playwright
    • about 635 unit tests
    • deployed on Vercel
    • accessibility scan in CI

    Deep diveLive siteRepo

  • 770+GitHub stars on a training program run in public

    Open-source program

    Run a training program in public whose repositories have 770+ stars and 130+ forks.

    Every graduate leaves with a public contribution history an employer can read.

    • Open source
    • Technical leadership
    • Mentoring
    • 26 community pull requests approved
    • 52 forks of Prework
    • $0 tuition

    Deep diveGitHub organizationCatalog

Numbers were counted from repository history, CI configuration and the GitHub API on 10 November 2026. Each deep dive names its sources.