⚡ FlashApply

AI-powered CV personalisation — tailored to every job offer, every country.

Upload your CV · Paste a job URL · Pick a country · Get a perfectly matched CV in seconds.

---

🗺️ Table of Contents

• What is FlashApply?
• Features
• Interface
• Scores explained
• Premium model
• Architecture
• Project structure
• Components
• External services
• Data contracts

---

🎯 What is FlashApply?

FlashApply is a Streamlit application that helps you adapt your CV to any job offer — automatically.

It analyses the gap between your CV and a target role, rewrites it using an LLM, and ensures the result conforms to the hiring norms of the target country. The full PDF export is unlocked via a one-time Stripe payment. Everything else is free to preview.

---

✨ Features

	Feature	Description
📄	CV Import	Upload PDF, DOCX, TXT, or image — OCR supported
🔗	Job Offer Parsing	Paste any job URL — text is extracted automatically via Jina AI
🌍	Country Targeting	Adapts formatting and tone for France 🇫🇷, Germany 🇩🇪, Canada 🇨🇦
📊	Match Scoring	Before/after keyword match score + country compliance score
🤖	LLM Rewriting	CV regenerated with DeepSeek-V3.2 reasoning (thinking) mode
💳	Stripe Premium	PDF download unlocked after a single payment — no account needed

---

🖥️ Interface

The Streamlit UI is split into two panels:

┌───────────────────────────┬─────────────────────────────────┐
│         LEFT PANEL        │          RIGHT PANEL            │
│                           │                                 │
│  ① Upload CV              │  ④ Match Original score         │
│  ② Paste job offer URL    │  ⑤ Match Custom score           │
│  ③ Select target country  │  ⑥ Conformité score             │
│                           │  ⑦ Personalised CV preview      │
│  [ Run Analysis ]         │  ⑧ PDF download  (premium 🔒)   │
│  [ Generate Stripe Link ] │                                 │
└───────────────────────────┴─────────────────────────────────┘

Supported CV formats: PDF · DOCX · TXT · Image (OCR)

---

📊 Scores Explained

Match Original — how good is your CV before personalisation?

Keywords are extracted from the job offer. This score measures how many of them already appear in your original CV. A low score means the offer's vocabulary is barely reflected in your current document.

Match Custom — how good is your CV after personalisation?

Same calculation, run on the rewritten CV. A higher score than Match Original confirms the LLM successfully aligned your CV with the offer's language and expectations.

Conformité — does your CV fit the target country?

This score does not compare your CV to the job offer. It checks against cultural rules specific to the selected country: section order, personal information policies, tone, and layout expectations.

Scoring formula:

match_score = (keywords found in CV ÷ total keywords detected) × 100

---

💳 Premium Model

┌────────────────────────────────────────────────────────────┐
│  FREE tier                      PREMIUM tier               │
│                                                            │
│  ✅ Upload CV                   ✅ Everything in FREE      │
│  ✅ Fetch & parse job offer     ✅ Download final PDF       │
│  ✅ Run full analysis                                      │
│  ✅ View personalised CV                                   │
│  ✅ See all three scores                                   │
│  ❌ Download PDF export                                    │
└────────────────────────────────────────────────────────────┘

Payment flow:

User clicks "Generate Stripe Link"
        ↓
Completes payment on Stripe checkout
        ↓
Stripe redirects → app/?payment=success
        ↓
App activates premium mode
        ↓
PDF download button appears ✅

---

🏗️ Architecture

Pipeline overview

flowchart TD
    U[Utilisateur] --> UI[Streamlit UI]
    UI --> P[Pipeline Agent]
    P --> CVX[CV Extractor]
    P --> JOBX[Job Offer Extractor]
    P --> CR[Country Rules]
    P --> MA[Matching Agent]
    P --> CGA[CV Generator Agent]
    CGA --> LLM[LLM Provider]
    UI --> PDF[PDF Builder]
    UI --> STRIPE[Stripe Payments]

    CVX --> FILES[PDF · DOCX · TXT · Image]
    JOBX --> JINA[Jina AI Mirror]
    CR --> CFG[JSON Country Config]

Loading

📁 Project Structure

FlashApply/
├── streamlit_app.py                    # Main Streamlit UI & pipeline entrypoint
├── ARCHITECTURE.md
├── README.md
├── requirements.txt
│
└── app/
    ├── main.py                         # Minimal FastAPI backend (health check)
    │
    ├── config/
    │   └── cv_customizer_agent.json    # Country profiles — FR · DE · CA
    │
    ├── agents/
    │   ├── pipeline_agent.py           # Core business orchestration
    │   ├── matching_agent.py           # Keyword-based scoring
    │   └── cv_generator_agent.py       # LLM CV generation + rule-based fallback
    │
    ├── extractors/
    │   ├── cv_extractor.py             # PDF · DOCX · TXT · OCR dispatcher
    │   └── job_offer_extractor.py      # URL → text via Jina AI
    │
    └── utils/
        ├── country_rules.py            # Country norm adapter
        ├── llm.py                      # DeepSeek / OpenAI connector
        └── stripe_payments.py          # Stripe payment link creator

---

🧩 Components

🎨 Presentation — streamlit_app.py

The single entry point for all user interaction.

• Collects CV file, job offer URL, and target country
• Manages Streamlit session state across steps
• Triggers run_pipeline and renders all results
• Generates the premium PDF from the personalised markdown
• Creates a Stripe payment link and unlocks PDF download on ?payment=success

---

⚙️ Orchestration — pipeline_agent.py

The central application service. Chains every step in order:

extract CV text
    → fetch job offer
        → load country rules
            → score original CV
                → generate personalised CV
                    → score personalised CV
                        → compute compliance
                            → return result dict

---

📂 Extraction

Module	Input	Method
cv_extractor.py	.pdf	pdfplumber
cv_extractor.py	.docx	python-docx
cv_extractor.py	.txt	Direct read
cv_extractor.py	Image	pytesseract OCR
job_offer_extractor.py	Any URL	https://r.jina.ai/{url}

---

🤖 CV Generator — cv_generator_agent.py

Generates the personalised CV using:

• 📋 Country formatting rules
• 📈 Current match score
• 🔑 Present / missing keywords
• 📄 Source CV text + job offer text

Falls back to a rule-based mode if no LLM API key is available.

---

🌍 Country Profiles — cv_customizer_agent.json

Each profile defines document length, layout type, allowed/forbidden personal info, section order, tone, and cultural constraints.

Country	Key rules
🇫🇷 France	Formal tone · photo optional · structured sections
🇩🇪 Germany	Lebenslauf format · photo expected · strict chronology
🇨🇦 Canada	No photo · concise · skills-first · bilingual considerations

---

🔌 External Services

LLM — llm.py

Provider resolution order:

1. DeepSeek   →   DEEPSEEK_API_KEY      deepseek-reasoner  (V3.2 thinking mode)
2. OpenAI     →   OPENAI_API_KEY        fallback only

deepseek-reasoner is the DeepSeek-V3.2 "thinking" mode — extended chain-of-thought reasoning before generating the final CV. Custom base URL is supported via DEEPSEEK_BASE_URL (defaults to https://api.deepseek.com).

---

💳 Stripe — stripe_payments.py

Creates per-session, on-the-fly:

Product  →  Price  →  Payment Link  →  Redirect to ?payment=success

No stored products, no subscriptions — each session generates its own Stripe artefacts.

---

📦 Local Dependencies

Package	Used for
pdfplumber	PDF text extraction
python-docx	DOCX parsing
pytesseract	Image OCR
Pillow	Image pre-processing
reportlab	PDF generation
streamlit	UI framework
fastapi	Backend health endpoint
langchain-openai	LLM client (DeepSeek-compatible)
stripe	Payment link creation

---

🔄 Data Contracts

Pipeline input:

{
  "cv_file_path": "local_temp_path",   # Temporary path of uploaded CV
  "offer_url":    "https://...",        # Job offer URL
  "country":      "France"             # Target country profile
}

Pipeline output:

{
  "country":               "...",      # Selected country
  "country_rules":         { ... },    # Loaded formatting rules
  "cv_text":               "...",      # Extracted CV text
  "offer_text":            "...",      # Extracted job offer text
  "original_matching":     { ... },    # Score before personalisation
  "personalized_cv":       "...",      # LLM-generated CV
  "personalized_matching": { ... },    # Score after personalisation
  "compliance":            { ... }     # Country conformity score
}

---

FlashApply · AI CV Personalisation · Streamlit + DeepSeek + Stripe

Not financial or legal advice. CV quality depends on LLM output.