Yash Saha

Developed an Anomaly Detection Bot to identify fraud, waste, and abuse with a precision of 74% in procurement transactions while using a combination of tree-based algorithms (Isolation Forest) and auto-encoder neural networks. Created Risk Bot to assess the risk associated with suppliers by crawling for relevant news articles and using question-answering (Bert QA) and text classification tasks to extract and evaluate financial, operational, and ethical risk entities. Led multiple projects and created services called Instareview and Contract Discovery to summarize and extract meaningful information/clauses pertaining to legal contracts between two - parties.

Led a team of three people to analyze customer transactions data from Xoxoday’s e-commerce platform (Plum) to gather insights, give recommendations based on products and analyze the effects of any new product/marketing strategies on business.

Developed multivariate time series models to forecast total citation amounts at daily and hourly intervals 2 months into the future for 26 parking sites in heavy traffic areas of Copenhagen. Involved in Digital Route Planning: A service to guide parking enforcement officers to the right parking spaces in real-time to maximize revenue through issuing citations.

Leveraged OpenCV’s HAAR Cascade classifier to detect faces in images, then fine-tuned Resnet-50 to extract facial landmarks which were features to an ensemble learning model for analyzing emotions.

Taught professionals of varying backgrounds and experiences wanting to transition into Data Science, leveraging R.

Completed 800 hours of a full-time intensive course, earning 100 ECTS (European Credit Transfer and Accumulation System) equivalent to a master’s degree. This course employed the ‘Learn by doing' methodology and I worked on my own on 6 projects with real data and business cases.