Sep 2021 – May 2023 · 1 yr 8 mos

• Business Insights Generation Tool For E-Commerce Data.
• Tech : Time-Series Analysis, Anomaly Detection, Clustering, K-Means, Z-Score, DBSCAN, NLG.
• Built and Deploy a Tool which Generates all the Business Insights combing the Sales, Traffic and Customers Data. We have built different set of Algorithmic Trees which uses different ML methods and Business Rules to generate all the Insights which replaces an Analyst’s work. We have used DBSCAN, I-FOREST, Z-SCORE, K-Means Clustering and other methods for Insight Generation.

Apr 2021 – Aug 2021 · 4 mos · Bangalore Urban, Karnataka, India · Hybrid

• Content Moderation for Forum:
• Tech : NLP, Sentence Transformer, Python, Convolution Layer
• Built the moderation pipeline using deep learning model which can automatically moderate the content for the forum. This model restricts the negative content as per the business requirement to be posted on the forum. We have cleaned the entire data and build the dataset for model training including synthetic data generation. We have also built model with different set of approach and compare result with different Approaches which includes building the entire pipeline starting from getting the data and filter it by using moderation.
• Entire Search Pipeline for Question. (Text and Image Similarity Search)
• Tech : NLP, Flask, Sentence Embedding, Faiss, MongoDB, ResNet50, Clustering.
• Built the entire pipeline for question to question search. We takes the question and find the similar question from our database to map it with the corresponding Learning Objective. We have designed the entire structure and implemented the pipeline for search functionalities. It Took the entire mapped data and build FAISS model on top the embedding vector to search.

Jan 2021 – Apr 2021 · 3 mos · Remote

• Automatic Key Value pair extraction from a pdf Document:
• Tech : Semantic Segmentation, Pytesseract, Clustering, Silhouette Score.
• Extracting the key value pair text provided inside the tables with the help of deep learning and puts the bounding box around each word. it also builds the relationship between each key and different values for that key(column wise) present inside a table. it detects the table and all it s column automatically then finds all the key value pair and builds the relationship among all those key value pair.

Aug 2019 – Jan 2021 · 1 yr 5 mos · Greater Bengaluru Area

• Correcting AI (Improving detection accuracy using Representer Function):
• Improving the accuracy of a Object Detection model with the help of Representer Function Algorithm. We have built an algorithm on top of YOLO algorithm and Representer function Algorithm which can improve the correctness of the detection model. It includes modifying the YOLO V3 loss function and YOLO V3 optimisation function with its implementation. Built different object detection model for different dataset. Understanding of and implementation of complete YOLOV3 architecture, training, optimisation process with Representer function algorithm. Complete design and development of the framework to improve the accuracy of object detection model.
• Build and Explain Video Classification Model using 3D-Convolutional Neural Network:
• Built a Toolbox which opens the AI decisions of a Video Classifier which takes a video segment(for ex: playing a cricket shot or riding bicycle etc) and the model as an input and explains the prediction result. It also provides the contribution of each pixel of each frame towards the classification and tells what are the important region in the input which contributes more towards classification.
• Built different Video Classification models for the use. Proposed the idea of explaining video classification models. Understanding of different algorithms and methods used to explain video classification model and building of different video classification model for different datasets.

Apr 2017 – Aug 2019 · 2 yrs 4 mos · Bengaluru, Karnataka, India

• Working with WIPRO Research and Development team to build new and in demand products with the help of Machine Learning and AI.
• Misclassification Dectection and Correction in Deep learning Models:
• Built a Toolbox to show why a mis-classification occurs(Layer-wise visualization), where exactly the test image deviates, gets mis-classified and how can we overcome it by showing the Movement of the image from the starting layer to the end layer and show which filter plays the role for deviation(mis-classification) and also generates textual explanations for misclasification.
• Deep Learning Library to Generate LRP:
• Building a Deep Learning optimized Framework to generate heat-map(from LRP) which supports the model of any architecture and framework(currently supports TensorFlow, PyTorch, Keras). It includes the implementation of all the deep learning layer from scratch for ex: CNN, FC, ReLu, Pooling layers etc.
• Sentiment Analysis on Indian Languages: (Word2vec, CNN, LSTM, Fasttext):
• Classification of sentences into different categories such as positive, negative, and neutral on indian languages using sentiment analysis. Built the classifier model to classify the sentences in to different categories. Used Word2vec and Fasttext embedding for indian language and also handled
• unseen word problem. Used CNN and LSTM to build the model.
• Automation of Reimbursement Portal: (OCR, CNN, Text processing):
• Designed a pipeline for automated claim response system which automates the claim reimbursement process and helps the team to optimize response time by minimizing the manual intervention. Built complete preprocessing code and AI model to fetch the information from the inputs. Built the entire pipeline of the automation tool of claim reimbursement process.

Dec 2016 – Mar 2017 · 3 mos · Bengaluru, Karnataka, India

• Completed Training on Python, AI and different Machine learning Techniques .
• Building Pneumonia Detection Classifier:
• Built a Deep Learning Classifier by using Inception V3 network architecture to detect whether
• someone is sick or not from their chest X-ray or their radiology report which classifies takes the X-ray image as an input and classifies it in pneumonia positive or negative classes.