tool-cv-corpus 0.2.0

Schema-driven, LLM-assisted engine that compiles a graph-of-atoms career corpus into job-targeted CVs, resumes, and cover letters.

tool_cv_corpus

Schema-driven, LLM-assisted engine that compiles a graph-of-atoms career corpus into job-targeted CVs, resumes, and cover letters.

Why

Most resume tooling treats a CV as a document you edit. Every new target becomes a rewrite, the sourcing behind each bullet gets lost, and one-off prose mixes with data you wanted to re-use.

tool_cv_corpus inverts the model:

1. Your corpus is structured, sourced career data with provenance.
2. A target is a job posting you are applying to.
3. The engine resolves corpus against target, scores claims by fit, applies visibility rules, and emits a format-agnostic RenderedResume.
4. Pluggable renderers turn it into PDF (Typst), JSON Resume, DOCX, or HTML.

Pipeline

flowchart LR
    LI[LinkedIn export] --> ING[ingesters]
    MD[Markdown + YAML] --> ING
    GH[GitHub profile] --> ING
    OR[ORCID] --> ING
    ING --> CORPUS[(corpus<br/>graph of atoms)]
    CORPUS --> LOAD[loader]
    TARGET[target] --> LOAD
    LOAD --> GEN[generator + LLM cache]
    GEN --> RR[RenderedResume<br/>intermediate]
    RR --> TYPST[typst -> PDF]
    RR --> JR[JSON Resume]
    RR --> DOCX[DOCX]
    RR --> HTML[HTML]

Quick start

uv add tool-cv-corpus
cv-corpus doctor                       # verify install
cv-corpus init my_career               # scaffold a corpus
cv-corpus validate my_career           # run 11 checks
cv-corpus schema --out schemas/        # export JSON Schemas

Try the bundled synthetic example:

cv-corpus validate examples/corpus_jordan_taylor

Entity kinds

person, organization, role, project, achievement, skill, education, publication, artifact, testimonial, cover_letter_seed, target, source_doc. Plus Claim records layered on top for sourced assertions.

See docs/architecture/schema.md for the design invariants.

Plugins

Three entry-point groups discover renderers, ingesters, and LLM providers at install time:

• tool_cv_corpus.renderers - typst, json_resume, docx, html
• tool_cv_corpus.ingesters - markdown, linkedin_export, github_profile, orcid
• tool_cv_corpus.llm_providers - anthropic, openai (stub)

Write your own: see docs/plugin_authoring/index.md.

Configuration

Variable	Effect
CV_CORPUS_SOURCE_STORE	Override CAS root (default: platformdirs)
CV_CORPUS_MODEL	Override default LLM model
ANTHROPIC_API_KEY	Anthropic provider auth
OPENAI_API_KEY	OpenAI provider auth (stub)

Status

Version 0.1.0 ships the schema, CLI surface, validator with 11 ordered checks, the default plugin set, CI matrix across 3 OS x 2 Python versions, and trusted-publishing release wiring.

The claim-scoring and target-aware generation phases are stubs; contributions welcome.

Contributing

Read CONTRIBUTING.md and code_of_conduct.md first. Conventional commits and SSH-signed commits are required for maintainers; contributors can open PRs without signing and a maintainer will squash-merge.

License

Apache 2.0. See LICENSE.