Paste a job description and your resume (or upload a PDF) → get an instant fit score, salary range estimate, keywords you're missing, AI-rewritten bullet diff viewer, a tailored cover letter (3 tones), interview prep, a skills gap roadmap, LinkedIn optimizer, email templates, and a standalone resume health check - all in one click.

Built to answer a real question every applicant has: "How well does my resume actually match this job - and what do I do about it?"

🔗 Try it live → resume-job-fit-ai.streamlit.app

A real recorded run on the live app (no edits): an AI-internship JD scored against a real resume - 85/100 with matched/missing keywords.

• Google Gemini (free tier, no credit card needed) via the official google-genai Python SDK
• Model: gemini-2.5-flash-lite - the most reliable free-tier model (overridable via GEMINI_MODEL env var)
• Structured outputs - Pydantic schemas passed as Gemini's response_schema; the model returns clean, validated JSON every time
• Streamlit front end - deployed free on Streamlit Community Cloud
• pdfplumber for PDF text extraction
• XSS prevention - all Gemini-generated strings are passed through html.escape() before rendering with unsafe_allow_html
• Auto-retry - exponential backoff on transient 429 rate-limits and 5xx server errors (up to 3 attempts)
• Defensive error handling: missing/invalid key, rate limits, oversized input, malformed responses all show friendly messages

git clone https://github.com/syzayd/resume-job-fit-ai.git
cd resume-job-fit-ai

python -m venv venv
venv\Scripts\activate            # Windows
# source venv/bin/activate       # macOS / Linux

pip install -r requirements.txt

cp .env.example .env             # paste your FREE key from aistudio.google.com/apikey
streamlit run app.py

Then click Load sample → Analyze fit → Generate all sections ✨ to see everything work instantly.

1. Fork this repo on GitHub.
2. Go to share.streamlit.io → New app → select your fork → app.py.
3. Under Advanced settings → Secrets, paste:

   GEMINI_API_KEY = "your-key-here"

   Get a free key (no credit card) at aistudio.google.com/apikey.
4. Click Deploy. Live in ~60 seconds.

resume-job-fit-ai/
├── app.py                        # Streamlit UI - 7 tabs, Generate All, PDF upload
├── analyzer.py                   # All Gemini logic - schemas, prompts, retry, error handling
├── requirements.txt
├── .env.example                  # GEMINI_API_KEY=your-key-here  (never commit .env)
├── .gitignore
├── README.md
├── .streamlit/
│   ├── config.toml               # Theme + server settings
│   └── secrets.toml.example      # Format for Streamlit Cloud secrets
├── .github/workflows/ci.yml      # GitHub Actions CI (pytest on every push)
├── pages/
│   ├── 1_Compare_Jobs.py         # Multi-job comparison page
│   ├── 2_Job_Tracker.py          # Application tracker page
│   └── 3_Company_Research.py     # Company culture, interview format, and prep tips
├── tests/
│   ├── test_analyzer.py          # 26 unit tests (Gemini mocked)
│   └── test_secrets_bridge.py    # 3 unit tests (Streamlit secrets shim)
├── db.py                         # SQLite persistence layer
├── docs/
│   └── screenshot.png
├── logs/                         # Build logs per path
├── sample/
│   ├── sample_resume.txt
│   └── sample_job.txt
└── handoffs/                     # Session handoff documents

Structured outputs are a multiplier. Passing a Pydantic schema as Gemini's response_schema turns the model from "hope it returns valid JSON" into a reliable typed component. No brittle string parsing - the SDK validates the response against the schema on every call.

Prompt design matters more than model size. The single most impactful instruction was "never invent experience the candidate doesn't have." It's one sentence, but it's what makes the rewrites actually trustworthy and usable. Quality of instruction beats size of model.

Good error handling is a feature. Most of the polish was making every failure surface as a friendly, actionable message - bad API key, rate limit, empty input, oversized input, PDF parse failure - instead of a stack trace. Users see this first, not the happy path.

Free-tier quirks are real constraints. gemini-2.0-flash has 0 free-tier quota right now. gemini-2.5-flash 503s under load. gemini-2.5-flash-lite is the actual reliable free-tier choice - you only know this by hitting failures in production.

Streamlit secrets ≠ env vars on Cloud. Streamlit Community Cloud injects secrets via st.secrets, not os.environ. A one-time shim at app startup (os.environ[k] = st.secrets[k]) bridges the gap cleanly without coupling the core logic to Streamlit.

Streamlit session state needs intentional keying. Streamlit reruns the entire script on every widget interaction. Without tracking pdf_name in st.session_state, the app re-extracted the PDF on every keypress. One extra state key eliminated the problem entirely.

-> NoReturn is not optional for always-raise functions. If a function always raises, annotating it -> None breaks type checker flow analysis - callers after _handle_api_error(exc) appear reachable. -> NoReturn + raise _handle_api_error(exc) at the call site is the correct pattern.

• PDF resume upload
• Cover letter generator
• Interview prep tab
• Skills gap roadmap
• LinkedIn profile optimizer (headline + About + skills)
• Copy-to-clipboard for cover letter and interview answers
• "Generate all sections" one-click button
• Streamlit Community Cloud deploy support
• Multi-job comparison (rank 2–3 jobs against your resume)
• DOCX export (formatted Word document)
• Tests + GitHub Actions CI
• Job application tracker (SQLite - save, track status, export CSV)
• Salary range estimator (pulled from JD, shown below score)
• Bullet diff viewer (side-by-side original vs rewrite, per-bullet copy)
• Cover letter tone selector (Professional / Warm / Bold)
• Score trend chart in job tracker (Altair line chart, 3+ entries)
• Analytics dashboard in job tracker (pipeline funnel + score histogram)
• Email templates (follow-up, thank-you, rejection response)
• Resume Health check (standalone quality score + quick fixes)
• Company Research page (culture, interview format, red flags, prep tips)
• Tailored Resume export (.docx with AI rewrites + ATS notes pre-filled)

Built in public by Zaid Ali Syed · github.com/syzayd Rewrites stay truthful to your resume - review before using.