AI-powered workflow that automatically classifies recruiter emails, checks if your resume aligns with the job description using semantic similarity, and drafts a reply with your resume attached. Built using Gmail API, a fine-tuned DistilBERT classifier, and SentenceTransformer embeddings.

• 🤖 AI-Powered Classification
  Fine-tuned DistilBERT model classifies recruiter emails with over 90% accuracy.

• 🧠 Resume Matching with Job Description
  Compares your full resume content against the recruiter's email/job description using semantic similarity (via SentenceTransformers). Replies are only drafted if the match is strong enough.

• 📧 Draft Replies
  Creates polite draft replies using a pre-written message template — with your resume (resume.pdf) attached.

• 🔄 Gmail API Integration
  Reads unread emails, processes them, and marks them as read using the Gmail API.

• 📜 Logging for Transparency
  Logs detailed messages for each step of the pipeline to help with debugging and transparency.

python mail_automation.py

2025-07-14 16:14:46,759 - INFO - Load pretrained SentenceTransformer: all-MiniLM-L6-v2
2025-07-14 16:14:50,753 - INFO - file_cache is only supported with oauth2client<4.0.0
2025-07-14 16:14:51,484 - INFO - Loaded fine-tuned model from ./fine_tuned_model
2025-07-14 16:14:51,754 - INFO - Found 2 unread emails
2025-07-14 16:14:51,914 - INFO - Classification result: non-recruiter for 'You have 1 new message'
2025-07-14 16:14:51,914 - INFO - Skipping NON-RECRUITER email: You have 1 new message
2025-07-14 16:14:52,464 - INFO - Classification result: recruiter for 'Software Development Engineer in Test || Location: San Diago/Carlsbad, CA || Only W2'
Batches: 100%|██████████████████████████████████████████| 1/1 [00:00<00:00,  9.33it/s]
Batches: 100%|██████████████████████████████████████████| 1/1 [00:00<00:00, 23.04it/s]
2025-07-14 16:14:52,685 - INFO - Resume-Job semantic similarity score: 0.609
2025-07-14 16:14:52,685 - INFO - High semantic match – creating draft for RECRUITER email: Software Development Engineer in Test || Location: San Diago/Carlsbad, CA || Only W2
2025-07-14 16:14:53,537 - INFO - Created draft for RECRUITER email: Software Development Engineer in Test || Location: San Diago/Carlsbad, CA || Only W2

• Python 3.8+
• Hugging Face Transformers
• PyTorch (CPU-compatible)
• Gmail API (Google Cloud)
• SentenceTransformers (all-MiniLM-L6-v2) for resume-job similarity
• scikit-learn (for evaluation metrics)
• PyMuPDF (for PDF resume parsing)

MailAIWorkflow/
├── assets/
│   └── resume.pdf                 
├── credentials/
│   ├── credentials.json          # Google OAuth
│   └── token.json                # Generated token
├── src/
│   ├── auth.py                   # Gmail OAuth
│   ├── classifier.py             # Email subject classifier
│   ├── gmail_client.py           # Gmail API utilities
│   ├── mail_automation.py        # Main script
│   ├── fine_tuned_model/         # Fine-tuned DistilBERT weights
│   ├── semantic_matcher.py       # Resume-job matching 
│   └── skills_matcher.py          
├── requirements.txt              # Python dependencies
└── README.md

The DistilBERT model was fine-tuned on a labeled dataset of recruiter and non-recruiter email subjects.

• Dataset:
  100+ manually labeled subjects (50/50 recruiter/non-recruiter)
• Accuracy: 92% on the validation set
• Training Time: ~15 minutes (on Intel UHD 620 CPU)

Note: The email classification model has been fine-tuned specifically for identifying recruiter emails related to Software Development Engineer in Test (SDET), QA Automation Engineer, or similar roles.

Want to adapt this project for your specific job role. You can fine-tune the classifier with your own labeled email subjects.

Step 1: Prepare Your Dataset Create a simple CSV file with two columns:

subject,label
"Urgent: SDET Position at Google",1
"Your Monthly Newsletter from ABC",0

You can start with as few as 100–500 rows.

Step 2: Run the Training

cd src
python train_model.py

What this does:

• Loads your dataset
• Fine-tunes DistilBERT on CPU
• Logs training progress and accuracy
• Saves the final model to ./fine_tuned_model/

Step 3: Use Your New Model Update mail_automation.py to load your fine-tuned model:

classifier = Classifier(fine_tune_path="./fine_tuned_model")

That’s it — your custom classifier is now live!

• Extracts full resume text from PDF (using PyMuPDF)
• Extracts recruiter job description from email body
• Generates semantic embeddings for both (using SentenceTransformers)
• Computes cosine similarity
• Only drafts reply if similarity score ≥ 0.5 (configurable)

1. Clone the Repository

git clone https://github.com/yourusername/MailAIWorkflow.git
cd MailAIWorkflow

2. Set Up Virtual Environment

python -m venv venv

Linux/Mac

source venv/bin/activate

Windows

venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Gmail API Setup

• Enable Gmail API in Google Cloud Console
• Download credentials.json into the /credentials/ folder
• Run the auth script to generate token.json:

cd src
python auth.py

5. Add Your Resume Place your resume as a PDF at:

assets/resume.pdf

cd src
python mail_automation.py

I'd love to hear your thoughts! What workflows have you automated recently? How would you improve this one? Feel free to open issues or share suggestions.

👤 Tasnim Fariyah 🔗 Github 🔗 LinkedIn