🔒 LockedIn

One resume per role, built from your real work.

A terminal app that reads your GitHub and your CV, then writes a tailored, ATS-friendly LaTeX resume for every internship you apply to. It uses the Claude Code or Codex CLI you're already logged into.

---

Why

Most students apply to AI, Backend, Frontend, DevOps and Data internships with the same resume, and recruiters and ATS filters notice. Tailoring by hand for every application takes hours, so nobody does it.

LockedIn does it in about a minute, using only things you've actually built.

What it does

1. Reads your GitHub. Repos, languages, topics, READMEs, and pinned projects.
2. Imports your CV (PDF / DOCX / TXT) and turns it into a clean LaTeX resume (Jake's Resume template) that opens straight in Overleaf.
3. Tailor mode. Pick a role, or paste the actual job description. LockedIn then:
   • ranks your repos for that role and features the strongest 3–4;
   • rewrites your bullets around the role's keywords;
   • scores the result with an ATS score, before → after;
   • fits it to one page and saves .tex + .pdf.
4. Never lies for you. A deterministic truth guard removes any project, skill, technology or number that isn't backed by your CV or repos.

Home	Pick a role (or paste a JD)
Saved, compiled, scored	The output (sample: fictional student, Backend role)

---

Installation

1. Prerequisites

Requirement	Why	How
Node.js 22+	Runs LockedIn	brew install node or nodejs.org / nvm
Claude Code or Codex CLI, logged in	Does the writing. LockedIn has no API keys of its own	Claude: npm i -g @anthropic-ai/claude-code, then run claude once and log in Codex: npm i -g @openai/codex, then codex login
GitHub CLI (recommended)	Reads your repos without you pasting a token	brew install gh, then gh auth login
Tectonic (optional)	Builds the PDF locally and checks it's one page	brew install tectonic (other OSes)

No gh? LockedIn will ask for a personal access token instead. No Tectonic? You still get the .tex file and one-key Open in Overleaf.

2. Install LockedIn

git clone https://github.com/anugrah55/LockedIn.git
cd LockedIn
npm install
npm run build
npm link        # makes the `lockedin` command available everywhere

Check it worked:

lockedin --version

Updating: git pull && npm install && npm run build. The linked command picks up changes automatically. Uninstalling: npm unlink -g lockedin, then delete ~/.lockedin if you want your data gone too.

3. First run

lockedin

Setup takes about two minutes:

1. Connect your AI. LockedIn detects Claude Code / Codex automatically.
2. Link GitHub. Choose public repos only (default) or include private ones.
3. Upload your CV. Drag the PDF into the terminal and press enter. Check the parsed summary, and you're done.

---

Usage

lockedin                          # home menu (or setup on first run)
lockedin tailor                   # pick a role → optionally paste the JD → review → save
lockedin tailor -r backend        # skip the role picker
lockedin tailor -r genai --jd jd.txt
lockedin sync                     # re-fetch your GitHub repos
lockedin cv ~/Downloads/cv.pdf    # re-import an updated CV
lockedin config                   # switch Claude/Codex, include/exclude private repos
lockedin roles                    # list built-in roles
lockedin where                    # show where your data lives

On the review screen you can:

• accept the draft;
• swap in different projects;
• regenerate;
• read what the truth guard removed.

Every version is kept in ~/.lockedin/outputs/<role>-<date>.{tex,pdf,json}. The .json includes the ATS report.

Built-in roles

✦ AI / ML Engineer · ✦ GenAI / LLM Engineer · ✦ Backend · Frontend · ✦ Full-Stack · ✦ DevOps / Cloud · Data Engineer · Data Scientist / Analyst · Mobile (Android / Flutter) · Cybersecurity · Blockchain / Web3 · ✦ SDE. You can also type any custom role.

_{✦ = hot for CS interns right now}

---

How it works

 ┌──────────┐   ┌──────────┐
 │  GitHub  │   │    CV    │   PDF / DOCX → text + hyperlinks
 └────┬─────┘   └────┬─────┘
      │ GraphQL      │ AI parse (strict JSON schema)
      ▼              ▼
   repos.json    profile.json          ← your data, stored locally in ~/.lockedin
      │              │
      └──────┬───────┘
             ▼
   ┌───────────────────┐   in parallel:  rank repos for the role
   │  analyze (AI)     │                 extract ATS keywords from the JD
   └─────────┬─────────┘
             ▼
   ┌───────────────────┐   select · reorder · reword. Never invent.
   │  tailor (AI)      │
   └─────────┬─────────┘
             ▼
   ┌───────────────────┐   drops unbacked projects, tech, skills, numbers
   │  truth guard      │   (plain code, not AI)
   └─────────┬─────────┘
             ▼
   ┌───────────────────┐   LaTeX render → Tectonic → shrink until 1 page
   │  render + ATS     │   ATS score before → after
   └─────────┬─────────┘
             ▼
     .tex · .pdf · Open in Overleaf

All AI calls go through your own claude -p or codex exec, with a strict JSON schema, no tools, and a throwaway working directory. LockedIn never sees or stores an API key.

ATS score

Part	Weight	What it measures
Keywords	70	Target keywords found in your resume. Required keywords count double. Spelling variants match ("NodeJS" = "Node.js", "Postgres" = "PostgreSQL").
Impact	15	Bullets that open with a strong action verb, and bullets with real numbers.
Structure	15	Contact info and core sections, no first person, no overlong bullets, one page.

Where the keywords come from:

• With a pasted job description, the AI extracts them from the JD.
• Without one, the role's curated keyword list is used.

The score itself is plain code, so the same resume always gets the same score. Missing keywords are shown, never added for you. Add them only if you've really used them.

The truth guard

The AI may only select, reorder and reword what's in your CV and repos. Then plain code checks its output:

• every project must map to a real repo or CV entry;
• a technology may appear on a project only if that project's README, metadata or CV entry shows it;
• skills must appear somewhere in your CV or GitHub. Concrete tools count as proof of concepts: Razorpay proves "payment gateway", GitHub Actions proves "CI/CD";
• rewritten bullets can't introduce numbers that aren't in the source.

Anything that fails is removed and listed on the review screen. The guard doesn't parse technologies named inside bullet prose, which is why you always review before saving.

Privacy

• Data stays local: everything lives in ~/.lockedin/ (override with LOCKEDIN_HOME).
• What leaves your machine: CV text and repo READMEs, sent to the AI provider you chose (Anthropic or OpenAI) via its official CLI.
• Private repos are opt-in, because their READMEs would be sent to your AI provider.
• Contact details (phone, email) are never sent during tailoring.

---

Troubleshooting

Problem	Fix
"LockedIn needs Claude Code or Codex"	Install one and log in: run claude once, or codex login.
Codex fails with 401 Unauthorized	codex login, then lockedin config → switch provider.
"Couldn't find a GitHub login"	gh auth login, or paste a token when asked.
"Could not read text from this CV"	Your PDF is probably a scanned image. Export a text-based PDF from Word / Google Docs / Overleaf.
Parsed CV has a small mistake	Edit ~/.lockedin/profile.json directly, or re-run lockedin cv <file>.
No PDF, only .tex	Install Tectonic (brew install tectonic), or press Open in Overleaf.

Development

npm install
npm run dev -- tailor     # run from source with tsx
npm test                  # vitest: escaping, golden LaTeX, guard, ATS, schemas, TUI
npm run typecheck
npm run smoke             # one real call to each logged-in AI CLI
npm run build             # bundle → dist/cli.js

src/
  cli.tsx            command-line entry (commander)
  app/               Ink screens: App router, onboarding steps, Tailor flow, UI kit, graphics
  ats/               keyword extraction + deterministic ATS scoring
  providers/         claude -p / codex exec adapters, strict JSON-schema output
  github/            token resolution (gh → env → PAT) + GraphQL repo fetch
  cv/                PDF/DOCX text + link extraction, AI parse → profile.json
  tailor/            roles, ranking, tailoring prompt, truth guard
  latex/             escaping + Jake's Resume renderer
  output/            Overleaf hand-off, Tectonic compile + one-page fit, saving
  schemas/           zod schemas: one source for validation and JSON schema
scripts/
  e2e.ts             headless full pipeline (CV → GitHub → rank → tailor → PDF)
  shots/             regenerate README screenshots from real app frames

Regenerating the screenshots

npm run build
LOCKEDIN_HOME=~/.lockedin-demo expect scripts/shots/capture.exp /tmp/shots ./jd.txt
for s in onboarding home roles review done; do
  npx tsx scripts/shots/term-shot.ts /tmp/shots/$s.log /tmp/shots/$s.svg
  scripts/shots/to-png.sh /tmp/shots/$s.svg docs/screenshots/$s.png
done

Roadmap

• Cover letter and cold-email generator from the same profile
• lockedin log: track where you applied, and with which version
• More LaTeX templates
• GitHub OAuth device flow (no gh or token needed)
• Gemini CLI provider
• Publish to npm

Credits

• Resume template: Jake's Resume by Jake Gutierrez (MIT).
• Built with Ink, zod, unpdf and Tectonic.

License

MIT © Anugrah