Smart draft guide

Practical guides to install and use the app effectively.

Smart Draft — AI-Powered Application Generation

Smart Draft transforms job descriptions into personalized application emails using advanced AI. This guide covers the complete workflow, supported providers, optimization techniques, and quality assurance.

How Smart Draft Works

┌─────────────────────────────────────────────────────────────────┐
│                    SMART DRAFT ARCHITECTURE                    │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  INPUT LAYER                                                    │
│  ├── Job Description (full text)                               │
│  ├── Company Context (name, website, industry)                 │
│  └── CV Selection (auto-matched or manual)                    │
│                          │                                      │
│                          ▼                                      │
│  ┌─────────────────────────────────────────────────────────┐   │
│  │              INTELLIGENCE LAYER                         │   │
│  │  ┌─────────────────┐  ┌─────────────────┐              │   │
│  │  │ Language        │  │ Tone Detection  │              │   │
│  │  │ Detection       │  │                 │              │   │
│  │  │ (langid.py)     │  │ (rule-based)    │              │   │
│  │  └─────────────────┘  └─────────────────┘              │   │
│  │  ┌─────────────────┐  ┌─────────────────┐              │   │
│  │  │ Entity          │  │ CV Matching     │              │   │
│  │  │ Extraction      │  │ Algorithm       │              │   │
│  │  │ (spaCy NER)     │  │ (TF-IDF + NLP)  │              │   │
│  │  └─────────────────┘  └─────────────────┘              │   │
│  └─────────────────────────────────────────────────────────┘   │
│                          │                                      │
│                          ▼                                      │
│  PROMPT ENGINEERING                                             │
│  ├── System Instructions (tone, language, format)               │
│  ├── Context Injection (CV highlights)                        │
│  └── Constraints (length, style, requirements)              │
│                          │                                      │
│                          ▼                                      │
│  AI EXECUTION                                                   │
│  ├── Provider Selection (DeepSeek/OpenAI/Anthropic)            │
│  ├── Model Routing (complexity-based)                         │
│  └── Response Generation                                      │
│                          │                                      │
│                          ▼                                      │
│  OUTPUT                                                         │
│  ├── Subject Line                                             │
│  ├── Personalized Body                                        │
│  └── Confidence Score                                         │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Supported AI Providers (15+)

Standard API Providers

Provider	Models	Best For	Cost
DeepSeek	deepseek-chat, deepseek-reasoner	Cost-effective, strong reasoning	$
OpenAI	gpt-4o, gpt-4o-mini, o3-mini	Industry leader, high quality	$$$
Anthropic	claude-3.5-sonnet, claude-3.5-haiku	Long context, nuanced output	$$
Google	gemini-2.0-flash, gemini-2.0-pro	Fast inference, multilingual	$$
Groq	llama-3.3-70b, mixtral-8x7b	Ultra-fast (< 100ms)	$
Mistral	mistral-large, mistral-medium	European data residency	$$
xAI	grok-2	Alternative perspective	$$
Cohere	command-r, command-r-plus	Enterprise features	$$
AI21	jamba-1.5-large	Long document processing	$$
Perplexity	sonar, sonar-pro	Research-backed outputs	$$

OAuth/Subscription Providers

Provider	Setup	Cost Model
Google OAuth	Link Google account	Existing Google AI access
GitHub Copilot	Copilot subscription	Included in Copilot Pro
ChatGPT Plus	OpenAI account	Included in Plus subscription
Claude Pro	Anthropic account	Included in Pro subscription

Local/Proxy Options

Option	Use Case	Setup
Ollama	Privacy-first, offline	Local LLM server
Gemini CLI Proxy	Google CLI integration	Terminal-based access
Antigravity Proxy	Custom proxy endpoint	Advanced users

Provider Configuration

Setting Up API Keys

Navigate to Settings → AI Providers
Select your preferred provider
Enter API key (stored in system keyring)
Select default model
Set cost limits and alerts

Model Selection Strategy

Use Case	Recommended Model	Why
Standard applications	deepseek-chat / gpt-4o-mini	Balanced cost/quality
Executive/C-level roles	claude-3.5-sonnet	Nuanced, sophisticated
Creative industries	gemini-2.0-flash	Dynamic, engaging
Research-heavy roles	o3-mini / sonar-pro	Analytical depth
High-volume campaigns	groq-llama-3.3-70b	Speed, cost efficiency

The Auto-Detection Engine

Language Detection

Smart Draft automatically detects the job posting language:

# Example detection flow
job_description → langid.classify() → confidence_score
                                    ↓
                           if confidence > 0.8:
                               use_detected_language
                           else:
                               fallback_to_cv_language

Supported Languages:

English (en) — 99.2% accuracy
French (fr) — 98.7% accuracy
Spanish (es) — 97.9% accuracy
German (de) — 97.5% accuracy
Arabic (ar) — 96.8% accuracy
Italian (it) — 97.1% accuracy
Portuguese (pt) — 96.9% accuracy
Dutch (nl) — 95.4% accuracy
And 100+ more

Tone Analysis

The system analyzes job description tone via keyword matching:

Tone Indicator	Keywords	Suggested Tone
Formal	"corporate", "enterprise", "regulatory"	Formal
Startup	"fast-paced", "disruptive", "agile"	Friendly
Creative	"design", "innovation", "passion"	Enthusiastic
Traditional	"established", "professional", "industry"	Professional

Entity Extraction

Smart Draft extracts key entities using spaCy NER:

ORG (Organization): Company names
POSITION: Job titles, role names
GPE (Geopolitical Entity): Locations
SKILL: Technologies, tools, methodologies
DATE: Posting dates, deadlines

Example:

Input: "Google is hiring a Senior ML Engineer in London"

Extracted:
- ORG: Google
- POSITION: Senior ML Engineer
- GPE: London

CV Matching Algorithm

Job Requirements
       │
       ├──→ Keyword Extraction (TF-IDF)
       │
       ├──→ Skill Matching
       │      ├── Exact match: weight 2.0
       │      ├── Related tech: weight 1.0
       │      └── Domain match: weight 1.5
       │
       ├──→ Experience Comparison
       │      ├── Years alignment: weight 1.5
       │      └── Level match: weight 2.0
       │
       └──→ Calculate Similarity Score
              │
              ▼
       Compare with all CVs
              │
              ▼
       Return Best Match + Confidence

Input Quality Optimization

Best Practices

DO:

✅ Paste the complete job description
✅ Include requirements section
✅ Add company context (mission, size)
✅ Mention specific technologies

DON'T:

❌ Use truncated descriptions
❌ Omit salary/location if relevant
❌ Copy only responsibilities (missing requirements)

Input Templates

Standard Job Posting:

Company: [Company Name]
Position: [Job Title]
Location: [City/Remote]

About the Role:
[Full description]

Requirements:
- [Requirement 1]
- [Requirement 2]

Tech Stack:
- [Technology 1]
- [Technology 2]

LinkedIn Job Posting:

Click "Show more" to expand full description
Copy entire text including "About the job" and "Qualifications"
Include "About the company" section

Tone Selection Guide

Professional (Default)

Best for: Most applications, balanced industries

Characteristics:

Clear, confident language
Balanced enthusiasm
Structured paragraphs
Appropriate formality

Example opening:

"I am writing to express my interest in the Senior Python Engineer position at [Company]. With 5 years of experience building scalable backend systems..."

Formal

Best for: Finance, legal, enterprise, government

Characteristics:

Structured, traditional format
Respectful deference
Precise language
Conservative tone

Example opening:

"I wish to submit my application for the position of Senior Python Engineer at [Company]. I am confident that my extensive experience..."

Friendly

Best for: Startups, creative agencies, small teams

Characteristics:

Warm, approachable
Conversational style
Shows personality
Enthusiastic but natural

Example opening:

"I was excited to come across the Senior Python Engineer role at [Company]! Your mission to simplify developer workflows really resonates with me..."

Enthusiastic

Best for: Dream companies, mission-driven roles, career transitions

Characteristics:

High energy
Strong alignment with mission
Passionate about opportunity
Forward-looking

Example opening:

"I've been following [Company]'s journey for years, and the Senior Python Engineer opening feels like the perfect convergence of my skills and my passion for..."

Review & Quality Assurance

The 4-Point Check

Before sending, verify:

Accuracy Check
- Job title matches the posting
- Company name spelled correctly
- Referenced skills are in your CV
- Dates and experience levels align
Tone Alignment
- Matches company culture
- Appropriate for industry
- Consistent throughout
Language Consistency
- Same language as job posting
- No mixed language fragments
- Correct grammar/spelling
Completeness
- Subject line present
- Opening hook engaging
- Body explains fit
- Call to action included

Common Issues & Fixes

Issue	Cause	Solution
Wrong company name	Poor entity extraction	Manual edit, re-generate
Mismatched skills	CV mismatch	Select correct CV, regenerate
Wrong language	Detection failure	Override language setting
Too generic	Vague job description	Add more company context
Too long	Complex requirements	Regenerate with "concise" prompt

Advanced Techniques

A/B Testing Templates

Generate multiple versions with different tones:

Generate draft with "Professional" tone
Save as Template A
Generate draft with "Friendly" tone
Save as Template B
Send to similar roles, track response rates
Identify winning approach for your industry

Custom Instructions

Add specific requirements to the prompt:

"Generate a cover letter that:
- Mentions my experience with [specific project]
- Highlights leadership in [specific context]
- Addresses the [specific requirement] directly
- Keeps to 3 paragraphs maximum"