Condense Cv

Condense CV

Read the master source first. In this repo, treat Latex/sample_cv.tex as the master CV and Latex/open-source.tex as a source fragment compiled into the master. Ignore out/ and generated PDFs unless the user explicitly asks about build artifacts.

Ask for the target page limit and audience if either is missing. Good defaults:

• 1 page for strict application portals and industry roles
• 2 pages for research, faculty, and senior technical roles

Prefer creating derived files instead of overwriting the master source. Default output paths:

• Latex/sample_cv_1page.tex
• Latex/sample_cv_2page.tex

Add a short header comment to each derived file stating the source file, target page budget, and intended audience.

If both page lengths are requested, derive both from the master source rather than deriving one from the other.

Cut content before tightening formatting. Use this order:

1. Remove entire low-priority sections.
2. Replace exhaustive lists with selected lists.
3. Merge repeated narrative into one-line entries.
4. Tighten spacing or typography only if the document is still slightly over budget.

Use references/selection-guide.md for page budgets, pruning order, and audience presets.

Keep these rules:

• Preserve factual accuracy. Do not invent claims, counts, or dates.
• Keep the contact block, current affiliation, and the most relevant research identity.
• Prefer recent, high-signal, role-relevant items over completeness.
• Keep dates, affiliations, named recognitions, and venue names intact.
• Favor selected publications, software, benchmarks, and systems over exhaustive lists.
• Remove teaching, service, student committees, talks, and older internships unless directly relevant.
• In this repo, prefer editing Latex/open-source.tex selections rather than treating Latex/open-source.json as the source of truth.

For a 1-page output:

• Keep contact, a 2-4 line summary, current role, education, and 2-4 strongest evidence sections.
• Limit publications to 2-4 items.
• Limit software/projects to 2-4 items.
• Keep awards or funding only if they materially strengthen the application.
• Prefer dense one-line entries and selected lists.

For a 2-page output:

• Keep contact, summary, employment, education, selected awards/funding, selected publications, and selected software/systems.
• Limit publications to 4-8 items.
• Limit software/systems to 3-6 items.
• Add service, teaching, advising, or talks only if the audience values them and space allows.

When the audience is specified, bias selection accordingly:

• academia: publications, grants, students, invited talks
• industry research / ML: systems, libraries, benchmarks, impact metrics, selected papers
• AI security / agent safety: agent-audit, Aegis, TrustLLM, agent security and auditing work
• anomaly detection / data mining: PyOD, ADBench, AD-AGENT, anomaly detection papers

If a LaTeX toolchain is available, compile and confirm the final page count. If compilation is unavailable, state that the source was prepared but the page count was not verified locally.