Ai-filtration Project lead, Back-end development

Ai filtration – is automated content filtration solution for social network which utilizes machine learning technology.

Ai-filtration works for following type of content:

• Nude/Adult content
• Still video uploaded
• Repeated video uploaded
• Reported content
• Mostly same color non meaningful photos
• Abusive audio
• Video without audio
• Curses in text or audio
• Text is extreeme or provocative
• Text is not in English/Spanish
• Text has abusive content

Back End is built in Python Django and following technologies

• Caffe - deep learning frameworks
• OpenCV - computer vision and image processing
• Google cloud speech API
• Pillow - image processing
• Celery - for multi-threading
• Libpuzzle - to find similar images
• Langdetect - detects language of the text
• psycopg2 - PostgreSQL adapter

This solution is built to receive hundreds of media and text posts and process them in real time. Django API receives post and transfers to celery for content automated filtration. After content is processed through all the filters, backend classifies the content and saves to database. Some posts are managed automatically. Some posts are reported to the administrator for further investigation.