Achraf MABROUK

- Designed and Developed a multi-file code conversion platform using Django, DRF, PostgreSQL, Redis, and Git; integrated three SAS code converters (built by the Data team). - Contributed to the development of a Generative AI-based system (OpenAI) for converting SAS scripts into Python and Alteryx workflows . - Collaborated with EY Paris to develop and maintain a tool for analyzing accounting data, utilizing Python, Pandas, Logging, and Tkinter. - Collaborated with EY Canada as a Backend Python Developer, playing a pivotal role in developing REST APIs using FastAPI, SQLAlchemy, and PostgreSQL.

- Conducted extensive research and analysis on cutting-edge text summarization models to develop an AI-based conversation summarization system - Extracted and prepared dataset including EDA and data cleaning steps - Contributed on the maintenance of the chatbot model with continuous data updates and data augmentation. - Managed model training, Docker image creation, and performed manual testing. - Contributed on transforming Python code of the data pipeline from Multi-class to multi-label classification. - Effectively collaborated with my AI team and the DEV team as well.

- Designed and developed a system that helps the employer to verify energy audit documents based on OCR and NLP. - Developed a Document Data Extraction (from PDF) system based on OCR. - Designed and developed a REST API using Django... - Developed Authentication system (including RBAC) - Built a machine learning model (NLP) to automatically recognize document type (0.83 accuracy). - Web Scraping: developed robot scripts to collect information from e-commerce websites and real estate program sites. - collaborated with Front-end developers..

Contributed to a project called "Cold Start Initialization": the project Intents to collect a massive product data sponsored by Influencers (eg. YouTube, Instagram) and feed them to Hippo app (to initialize the database). Eg. YouTube Crawler : Designed and developed a custom YouTube crawler that successfully crawled over 9K YouTube channels and extracted over 500K video data. - wrote a script to automatically create and deploy the crawler image in parallel on 9 DigitalOcean instances. - a Kafka broker is used to store the stream of data. - wrote a script to consume the data from Kafka broker and wrote them to JSON Lines file.

- Proposed an approach that automatically extracts specific information from scanned Invoices (images) with different layouts based on Fuzzy Text Matching and Rule Induction. - OCR engine is used at first place to convert images into machine-readable text. - Fuzzy text matching technique is used to detect label fields of the information to be extracted. - a set of rules (spatial features) is defined to identify the target data. *Main tasks: - Conducted an intensive research on document data extraction field - Feature engineering: Analysing document layouts and extracting textual and positional features. - Developed an algorithm that treats various document layouts. #NLP #Computer_Vision

Developed a tool allows you to Search, Download online videos and Extract Audio and Images from the downloaded videos. This tool is done for the sake of helping researchers to collect audio-visual data to use later for machine/deep learning models. 🛠 Programming language and APIs/Libraries used: ✔ Python 3 ✔ FFmpeg ✔ OpenCV ✔ PyQT5 ✔ SQLite ✔ Google Search API ✔ Bing Search API ✔ Youtube video downloader (youtube_dl)