Skip to content
Devs-GMTPublic

About

An AI CV checker and job matcher.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Callback logo

Callback

Get the callback.
Check how ATS-friendly your CV is, how well it matches a job, and exactly what to change —
then re-check it, export a clean CV and cover letter, and optionally stamp your progress on BOT Chain.

Live site Next.js 16 React 19 TypeScript Tailwind CSS v4 Gemini Solidity Deployed on Vercel

Live site · Check my CV · Sample report · Contract (testnet) · Privacy · Terms

The Callback home page: Get the callback.


Contents


Links

Live site https://callback-six-theta.vercel.app
Start a check https://callback-six-theta.vercel.app/app
Sample report https://callback-six-theta.vercel.app/#sample
My applications https://callback-six-theta.vercel.app/app/applications
Contract — BOT Chain testnet (Bohr, chain 968) 0xd949113c7AA4E2E9BF44bf5774D718211f4F38A5
Contract — BOT Chain mainnet (chain 677) not deployed yet — see Deploying the contract
Explorers Testnet https://scan.bohr.life · Mainnet https://scan.botchain.ai
Privacy policy / Terms /privacy · /terms
Product requirements PRD.md

Features

🤖 ATS check Seven weighted, rule-based checks with ✓ / ⚠ / ✗, the points earned and a one-line fix for each.
🎯 Job match with evidence Every requirement in the job is matched against your CV with a quote from your CV — checked in code, so a made-up quote loses the match.
✍️ What to pivot Rewrites in the job's language, built only from what your CV already says. Keywords you can't back up go in an "add only if true" card instead of into your CV.
📄 Improved CV Accept the changes you agree with and see them highlighted in place, then re-check and watch both scores move.
🎨 Five ATS-safe templates Professional, Modern, Classic, Compact and Elegant, in six accent colours, with a centred or left header — exported as a text-based PDF and a native Word .docx.
✉️ Cover letter A one-page letter in three lengths and three tones, every body paragraph backed by a line from your CV.
🛠️ CV editor Bold, italic, links and bullets from a toolbar (Ctrl+B / I / K), plus add, rename, reorder and hide sections, with "Read as" labels showing how an ATS files each one.
🔒 Private by design Files are parsed in your browser. Reports live in your browser. Nothing about you is stored on a server.
⛓️ Optional on-chain stamp Two hashes, two scores, a job title and a date on BOT Chain — proof of your progress over time, with no personal data.

New check: add your CV and the job


How it works

  1. You add a CV — upload a PDF or .docx, paste the text, or give a public link. Files are parsed in your browser; the file itself is never uploaded.
  2. You see what we read. Extraction from a PDF is never perfect, so the text is shown back to you to correct before anything is scored.
  3. You add the job — paste the description, or give the URL of a career page.
  4. You get a report with an ATS score, a job match percentage, every requirement checked with a quote from your own CV as evidence, and suggested rewrites.
  5. You accept the changes you agree with, edit the result, pick a template, re-check, and watch both scores move.
  6. You export the improved CV as a PDF or .docx, write a matching cover letter, download the report — and optionally stamp the result on BOT Chain.

CV templates and cover letter

Four of the CV templates: Professional, Modern, Classic and Elegant

Every template is single-column, with standard headings, real bullet lists and no tables, text boxes or images — the layout an applicant tracking system reads most reliably. Templates differ only in type, colour and spacing, so changing the look never costs the score.

  • One definition, three renderers. Each template is described once, in points, in src/lib/cv-templates.ts. The on-screen preview, the PDF (cv-pdf.ts) and the Word file (cv-docx.ts) all read the same numbers, so what you see is what downloads.
  • The PDF is real text, set in embedded Inter and Newsreader (both OFL), so it can be selected, searched and parsed. Heading letter-spacing is capped at 0.05 em: from about 0.09 em, PDF text extraction reads "EXPERIENCE" as "E X P E R I E N C E".
  • The CV is read into a structure first (cv-layout.ts): name, headline and contact details; sections; role / organisation / dates on one line with the dates on the right; the job title in italics underneath. Lines a PDF wrapped mid-sentence are joined back up.

A CV and its matching cover letter

The cover letter uses the CV's header and template so the two read as a set, laid out as a business letter: date, recipient, a bold subject line, greeting, an opening hook, evidence-led body paragraphs and a closing with a call to action. It comes in Short (260–300 words), Standard (340–400) and Detailed (440–500).

It follows the same no-invention rule as everything else: every body paragraph must quote a line from your CV, and the quote is checked in code (cover-letter.ts). A paragraph whose quote is not in your CV is dropped, and the quotes are shown under the letter as "Backed by your CV". It has its own daily limit, separate from analyses.


Editing the CV

The Improved CV tab has an editor and a Sections panel.

Action How
Bold / italic Select text, then B / I or Ctrl+B / Ctrl+I. Written as **bold** and *italic*.
Link Select text, then 🔗 or Ctrl+K. Written as [text](https://…). Emails, LinkedIn and GitHub in the header are linked automatically.
Bullets Select lines, then • to bullet or un-bullet them all.
New section from a line Put the cursor on a line and press H: it becomes a section heading. Press again to undo.
Add a section + Add section in the Sections panel: Projects, Certifications, Organizations, Awards, Volunteering, Languages, Courses, or a name of your own. The editor opens with a skeleton selected to type over.
Rename, reorder, hide ✏️, ↑ ↓ and 👁 on each section. Hidden sections stay in your text but are left out of the preview, downloads, copy and re-check.

Each section shows how an ATS will file it — Experience, Education, Skills, Summary (the sections the score counts) or Extra. Indonesian headings such as Pengalaman Kerja, Pendidikan, Keahlian and Ringkasan are recognised too.

Formatting markers are stripped before anything is scored, sent to the AI or copied, so formatting can never change a score. In the PDF and .docx they become real bold, italics and clickable links (inline.ts).


How the scores are calculated

Both numbers are computed in code, not by the AI, so the same inputs always produce the same output. The AI's job is to read and judge; the arithmetic is ours and is published here.

ATS score — 0 to 100

Check Points What it looks for
Text is readable 20 Enough extractable text. An image-only (scanned) PDF fails this outright.
Standard sections found 20 Experience, Education and Skills (6 each), plus a Summary (2).
Contact details present 10 Email (3), phone (3), city or location (2), LinkedIn or portfolio (2).
Dates are consistent 10 Dates present, in one dominant format, with real ranges rather than bare years.
No formatting risks 15 Starts at 15, then deducts: two columns −6, tables −4, images −3, details in a header/footer −2.
Length is right 10 1–2 pages full marks, 3 pages partial, 4+ fails. Estimated from word count when no page count is available.
Bullet quality 15 Bullets that start with a strong verb and carry a number or result. The one AI-assisted check; if the AI is unavailable, a rule-based fallback scores it from action verbs and quantified results.

Source: src/lib/analysis/ats.ts.

Job match — 0 to 100%

score = must-haves (60%) + nice-to-haves (20%) + keyword coverage (20%)
  • Each requirement is matched (full credit), partial (half credit) or missing (none).
  • Empty buckets are redistributed, not scored as zero. If a job has no nice-to-haves, that weight is spread across the buckets that do have content — otherwise a perfect CV would be capped at 80% for a reason that has nothing to do with the CV.
  • Keyword coverage is computed in code, case-insensitively, with a synonym list — "Postgres" covers "PostgreSQL", "JS" counts as "JavaScript".
  • A seniority mismatch is a warning, not a deduction.

Source: src/lib/analysis/match.ts, src/lib/analysis/keywords.ts.

Why you can trust the match

The prompt tells the model that a requirement only counts if it can quote the supporting line from the CV. That instruction is then enforced in code: every quote is checked against the CV text, and a claimed match whose quote is not actually there is demoted before it reaches your score. See verifyEvidence in match.ts.

The no-invention rule

Rewrites may only rephrase facts already in your CV. A keyword your CV does not support is never written in for you — it appears in an "add only if true" card, and accepting it changes nothing in the CV text. Suggestions never rewrite, move or delete a section heading. See prompts.ts and apply.ts.


Privacy

  • Files never leave your browser. PDF parsing uses pdf.js and .docx parsing uses mammoth, both client-side.
  • Your CV text is sent to one place: our API route, to make the AI call. It is not written to a database, a file or a log. Requests go to Gemini with store: false.
  • Reports live in your browser (localStorage), never on a server.
  • "Delete all my data" on the My applications page removes every report and the hash salt.
  • Rate limiting stores a counter per IP and nothing else — 8 analyses and 8 cover letters per IP per day.

What goes on-chain

Only if you press Stamp on BOT Chain:

Stored Not stored
Hash of the report Your name
Salted SHA-256 of your CV text Your email or phone
ATS score, match score Any CV text
Job title (≤ 60 bytes) The company name
Id of the earlier report it improved on Anything else
Timestamp, and the sending wallet address

The CV hash is salted with a random value generated in your browser that never leaves it, so nobody holding a copy of your CV can confirm it matches an on-chain record.

Honest caveat: the scores are computed by this app. A stamp proves your own progress over time — not that an employer agreed with the score, and it is not a credential.


The contract

contracts/contracts/Callback.sol, Solidity ^0.8.20. No tokens, no payments, no admin role, no upgrade path. Anyone can stamp their own reports; nobody can alter or remove them.

function stamp(
    bytes32 reportHash,
    bytes32 cvHash,
    uint8   atsScore,     // ≤ 100
    uint8   matchScore,   // ≤ 100
    uint256 improvedFrom, // 0, or an earlier report of yours
    string calldata jobTitle // ≤ 60 bytes
) external returns (uint256 id);

function getReport(uint256 id) external view returns (Report memory);
function reportsOf(address owner) external view returns (uint256[] memory);
function stats() external view returns (uint256 totalReports, uint256 totalRechecks);

Report ids start at 1: index 0 is an unusable placeholder owned by address(0), because improvedFrom == 0 means "a first check". improvedFrom must be an existing report owned by the caller, so you cannot claim someone else's score as your "before".

Network Chain ID RPC Explorer Address
BOT Chain testnet (Bohr) 968 https://rpc.bohr.life https://scan.bohr.life 0xd949…38A5
BOT Chain mainnet 677 https://rpc.botchain.ai https://scan.botchain.ai not deployed yet

Deploying the contract

Deployment is done by hand in Remix, so no private key is needed anywhere in this repo.

  1. Open Remix and paste in Callback.sol.
  2. Compile with Solidity 0.8.20 or later.
  3. Deploy to BOT Chain testnet first and stamp a test report and a re-check.
  4. Deploy to BOT Chain mainnet.
  5. Put the address in NEXT_PUBLIC_CONTRACT_ADDRESS (and remove NEXT_PUBLIC_USE_TESTNET for mainnet), then update the tables above.

Until the address is set, the stamping UI is disabled with an explanation rather than broken, and every other feature works.


Running it locally

Requires Node 22 or later.

git clone https://github.com/Devs-GMT/Callback.git
cd Callback
npm install
cp .env.example .env.local   # then add your Gemini key
npm run dev                  # http://localhost:3000
Script What it does
npm run dev Dev server (copies the pdf.js worker into public/ first)
npm run build Production build
npm run typecheck tsc --noEmit
npm run lint ESLint
npm test Unit tests — scoring, CV layout, formatting, cover letter, schemas, SSRF ranges
npm run contracts:test The Solidity test suite

Environment

See .env.example. No key is committed, and .env*.local is gitignored.

Variable Required Purpose
GEMINI_API_KEY For the AI features Free tier key from AI Studio
UPSTASH_REDIS_REST_URL / _TOKEN Recommended in production Makes the per-IP limit real on serverless
NEXT_PUBLIC_CONTRACT_ADDRESS For stamping Deployed Callback.sol address
NEXT_PUBLIC_USE_TESTNET=1 Optional Use BOT Chain testnet (968) instead of mainnet (677)
NEXT_PUBLIC_SITE_URL Optional Canonical origin for share links; inferred on Vercel
GEMINI_MODEL_ANALYZE / _PIVOT Optional Model overrides (default gemini-3.5-flash-lite)
GEMINI_THINKING_ANALYZE / _PIVOT Optional minimal · low · medium · high

Without a Gemini key the app still runs: the rule-based ATS check works in full, and the job match tab explains itself and offers a retry. Without Upstash the limit falls back to an in-process counter that resets on every cold start — fine locally, not a real limit on Vercel.


Deploying to Vercel

  1. Import the repository. Root directory: ./ (framework: Next.js).
  2. Add the environment variables above for Production and Preview.
  3. Deploy. prebuild copies the pdf.js worker automatically.
  4. After changing an environment variable, Redeploy so it takes effect.

The contracts/ directory is its own package and is not installed by the Vercel build. This app needs a server for its API routes, so it is not suited to static hosting such as GitHub Pages.


How it's built

Area Choice Why
Framework Next.js 16, App Router, React 19, TypeScript Cache Components on, so the marketing pages are static and the app shell prerenders
Styling Tailwind CSS v4 Design tokens live in @theme in globals.css; the CV preview scales with container query units
AI Gemini via @google/genai A genuine free tier, and a JSON Schema the decoder enforces. Behind one adapter (provider.ts), with a fallback when a schema is refused
Validation Zod v4 Every AI response is validated and repaired, so one odd field degrades a row rather than the report
Extraction pdf.js, mammoth Client-side, which keeps the privacy promise literal
CV & letter export jsPDF (embedded fonts) and docx Selectable vector text and native Word lists and headings, not a screenshot
Rate limiting Upstash Redis Separate daily budgets for analyses and cover letters
Chain ethers v6 MetaMask, chain add/switch, and read-only RPC for share links
Artwork Hand-authored SVG The logo, the handset mark and the torn-paper edge are drawn in this repo

Notable implementation details

  • Two-column PDFs are detected, not silently fixed. pdf.ts finds a vertical gutter, reads each column in order, and raises a warning, because an ATS would interleave the columns.
  • The link reader is fenced in. safe-fetch.ts validates every redirect hop and every DNS address, and allows only http/https on default ports.
  • An AI failure is a first-class outcome. A failed analysis still produces a saved report with the full rule-based ATS check and a retry.
  • Text copied from a PDF is handled as it really comes out: no blank lines between sections, sentences wrapped mid-line, names in capitals. The layout and the suggestion engine are tested against exactly that.

Project structure

src/
  app/                 Routes: website, /app, /app/report/[id], /api/fit/*
  components/
    site/              Marketing website sections
    app/               The app: flow, report tabs, CV editor, templates
    ui/                Buttons, cards, fields, gauges, tabs
  lib/
    ai/                Gemini adapter, prompts, response schemas
    analysis/          ATS, match, keywords, CV layout, formatting, cover letter
    export/            CV / cover letter PDF and .docx, report PDF
    extract/           PDF, .docx and link extraction
    chain/             Wallet and contract calls
    server/            HTTP helpers, rate limit, safe fetch
contracts/             Callback.sol and its Hardhat tests
tests/                 Unit tests (node:test via tsx)
docs/images/           README images
public/fonts/cv/       Inter and Newsreader for the PDF export (OFL)

Testing

npm test                 # app unit tests
npm run contracts:test   # Solidity tests
npm run typecheck && npm run lint && npm run build

The app tests cover the scoring arithmetic, evidence verification, CV layout (including PDF-extracted text), applying suggestions, inline formatting and the editor commands, section edits, cover-letter verification, the AI response schemas and the SSRF address ranges. The contract suite covers the id scheme, score and title bounds, the re-check ownership rules, the views, and the absence of any privileged entry point.


Scope

Built: ATS check, job match with evidence, pivot suggestions, improved CV with editor, five templates with PDF and .docx export, cover letter, report PDF, local history, and on-chain stamping.

Languages: English (default, unprefixed URLs), Indonesian (/id) and Malay (/ms) — the interface, reports, exports, AI feedback and legal pages are translated. The English legal text governs if a translation differs.

Not included, by design: auto-applying to jobs, scraping LinkedIn, recruiter accounts, server-side CV storage, payments, NFTs.


Disclaimer

Callback is an independent project and is not affiliated with, sponsored by or endorsed by any job board or ATS vendor. Names of third-party products are used to describe compatibility only and remain the trademarks of their respective owners. Scores are guidance, not a hiring guarantee. Your CV is never stored and never put on-chain.


Made for job seekers who deserve a callback.

About

An AI CV checker and job matcher.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages