Resume Optimizer

An AI-powered resume optimizer built on Anthropic's Claude API. Optimize your resume for ATS, score keyword coverage across dimensions, and walk into interviews prepared.

---

Features

Module	What it does
Resume Optimizer	Claude rewrites your resume to match the JD. ATS-friendly, keyword-targeted, role-toned. Streams output live, explains every change, flags hallucinations, rewrites weak bullets in STAR format, and exports to PDF.
ATS Scorer	Multi-dimensional scoring: keyword match, skills coverage, experience level, title alignment. Categorized gaps with actionable suggestions. Validated with Pydantic so the UI never breaks on bad output.
Interview Prep	10 role-specific questions (5 technical + 5 behavioral), STAR templates, talking points, red flags. Salary negotiation script for India/remote market.

What makes it production-ready

• PDF upload: upload your resume as a PDF instead of pasting text (pdfplumber, multi-column aware)
• STAR rewriter: scans for weak passive bullets ("responsible for", "worked on") and rewrites them in STAR format
• Hallucination check: flags any skills or tools Claude added that were not in your original resume or the JD
• Pydantic validation: all structured Claude responses are validated before hitting the UI
• Structured logging: every API call is logged with duration, token usage, and cost to logs/optimizer.log
• Input limits: resume capped at 5,000 chars, JD at 3,000 chars to prevent abuse
• Bring your own API key: paste your Anthropic key in the sidebar, never stored anywhere
• Graceful error handling: overloaded API, bad JSON, and parse failures all show clean messages

---

Architecture

Streamlit App
  app.py (home)
  pages/1_optimizer.py  ->  optimizer.py
  pages/2_ats.py        ->  ats.py
  pages/3_interview.py  ->  interview.py
              |
        prompt.py  (all Claude prompts as named constants)
        models.py  (Pydantic models for structured responses)
              |
       Anthropic Claude API  (claude-sonnet-4-5)

  pdf_builder.py   ->  fpdf2 (cross-platform, no system binaries)
  utils/sanitize.py -> input cleaning + PDF text extraction
  utils/logger.py  ->  rotating file logger
  utils/session.py ->  versions/*.json (session history)
  utils/cost.py    ->  token usage + cost display

---

Local Setup

Works on macOS, Linux, and Windows. No system binaries required.

1. Clone

git clone https://github.com/areychana/resume-optimizer.git
cd resume-optimizer

2. Create a virtual environment

python -m venv .venv

# macOS / Linux
source .venv/bin/activate

# Windows
.venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Set your API key

cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY

Get your key at console.anthropic.com.

Or just paste it in the sidebar when the app is running (bring your own key mode).

5. Run

streamlit run app.py

---

Running Tests

pytest tests/ -v

Tests cover: input sanitization, Pydantic ATS model validation, score clamping, fallback scoring, and PDF generation.

---

Deploy to Streamlit Cloud

1. Fork this repo
2. Go to share.streamlit.io and connect your fork
3. In Settings -> Secrets, add:

   ANTHROPIC_API_KEY = "sk-ant-..."

4. Deploy. No other configuration needed.

See .streamlit/secrets.toml.example for the secrets template.

---

Project Structure

resume-optimizer/
├── app.py                        # Home page
├── optimizer.py                  # Resume rewrite, STAR rewriter, hallucination check
├── ats.py                        # ATS scoring with Pydantic validation
├── interview.py                  # Interview prep + salary script
├── models.py                     # Pydantic response models
├── pdf_builder.py                # Markdown to PDF via fpdf2
├── prompt.py                     # All Claude prompts as named constants
├── pages/
│   ├── 1_optimizer.py            # Optimizer page
│   ├── 2_ats.py                  # ATS scorer page
│   └── 3_interview.py            # Interview prep page
├── utils/
│   ├── sanitize.py               # Input cleaning + PDF text extraction
│   ├── logger.py                 # Structured rotating file logger
│   ├── cost.py                   # Token usage + cost estimation
│   └── session.py                # Session history (JSON)
├── tests/
│   ├── test_sanitize.py          # Sanitizer edge case tests
│   ├── test_ats.py               # Pydantic model + fallback scoring tests
│   └── test_pdf.py               # PDF generation tests
├── versions/                     # Auto-saved session JSON files
├── logs/                         # Auto-generated API call logs (gitignored)
├── .streamlit/
│   ├── config.toml               # Theme + server config (10 MB upload limit)
│   └── secrets.toml.example      # Streamlit Cloud secrets template
├── .env.example                  # Local env template
├── CONTRIBUTING.md               # Contribution guide
└── requirements.txt              # Pinned dependencies

---

Tech Stack

• AI: Anthropic Claude API (claude-sonnet-4-5), streaming + structured JSON output
• UI: Streamlit multipage app with session state and live streaming
• Validation: Pydantic v2 for structured Claude responses
• PDF input: pdfplumber (multi-column aware extraction)
• PDF output: fpdf2 (pure Python, zero system dependencies)
• Logging: Python logging with RotatingFileHandler
• Storage: JSON files, no database
• Config: python-dotenv + Streamlit secrets

---

Cost

Each run costs approximately:

• Resume optimization: ~$0.005 to $0.015
• ATS scoring: ~$0.003 to $0.008
• Interview prep: ~$0.008 to $0.020

Costs are shown live in the UI after each generation. Users bring their own API key so there is no shared cost exposure.

---

Contributing

Contributions are welcome. See CONTRIBUTING.md to get started.

---

License

This project is licensed under the MIT License.