AI Email Scraper

At OFE, a maritime technology startup based in Singapore, I worked on an AI email scraper that automated the extraction of relevant information from emails and attachments.

From email to structured information

The workflow covered data retrieval, classification, extraction, and cleaning. Incoming emails were retrieved and stored as dictionaries in MongoDB. The classification stage distinguished between body text and attachments, and assigned the email text to categories needed by the product team.

For unstructured email text, we treated extraction as a named entity recognition problem, using an LSTM-CNN-CRF model implemented with TensorFlow.

Working with varied formats

Attachments included spreadsheets, PDFs, images, and plain text. Their inconsistent formats required extraction logic that could accommodate different layouts and structures.

This was a private project, so the source code and client data are not publicly available.