Machine Learning Engineer
As the part of the R&D team, I develop and implement solutions in the domains of Machine Learning. I have worked on projects for clients that include the leading consulting companies, and multinational banks of the world. Some notable projects are listed below.
- Risk Analysis
- Banks receive millions of documents for violating terms. Determining whether or not the text is pertinent requires one to read it in completion, this platform aims to ameliorate this problem by evaluating the document for possible risks and violations and predicting the extent of the potential loss for the bank.
- Bidirectional Long Short Term Memory Network, Word Embedding, and Convolutional Neural Network were used to classify the text for risk with a precision of 92% and an accuracy of 88%.
- On classification of a risk, an Extremely Randomized Trees classifier was used to predict the extent of risk (Very risky to less risky) hence enabling the bank to address the most relevant issues first.
- Technology stack: Python, Keras, NLTK, D3JS.
- Brand Sentiment Analysis
- The BSA platform helps companies to determine the success of their products in the market based on sentiment analysis on empirical and live stream data from sources like Twitter, Facebook, News portals, and other sources giving critical insights into how the end customer feels about the product.
- Bagged model with Naive Bayes, Max Entropy, and Multilayered Perceptron was trained with an accuracy of 83%
- Technology stack: Python, Flask, Sklearn, PyEnchant, GEvent, SocketIO, D3JS.
- Invoice Filter
- Extracting the required fields from scanned invoices is a challenging and tedious task, Invoice Filter aims to solve it.
- Pre-processing was expedited using Cython. Image processing techniques like Binarisation, Deskewing, Rescaling were achieved using OpenCV, and ImageMagick. Finally, a custom Tesseract model was trained for the OCR task.
- Technology stack: Python, Cython, Flask, NLTK, OpenCV, Tesseract.