Professional Projects :

Here are the projects I worked on while in software companies. I have worked with multiple clients from different countries. We were working on the problem statements which would make world to be a better place to live in. Some of the major projects are explained below.

Driver Drowsiness Level Detection in an Automobile

Skills : Computer Vision, Time Series, Deep Learning

- I was working on this problem statement in my company at Bosch Global Software Technologies.
- As we all know, a lot of accidents happen due to driver feeling drowsy if driving for a long time. It is important to work on this problem statement as to avoid these accidents. - Our aim was to detect the drowsiness level of driver which was our target feature. It was labelled as 1 - 10, 1 being least drowsy and vice-versa.
- We prepared our dataset using facial features extracted from images taken using interior cameras. Video data was collected through various. drivers and those videos were converted to base-signals using feature extraction techniques which were there used for data preparation and model training part.
- We framed this as a time series problem by drawing an analogy that a driver feels drowsy after driving it for a long time. We performed experiments using various deep learning algorithms (ANNs, RNNs, LSTMs).

Driver Seat-Belt Detection in an Automobile

Skills : Computer Vision, Deep Learning, Explainable AI

- I was working on this problem statement in my company at Bosch Global Software Technologies.
- A lot of critical injuries might happen if a driver meets with an accident if seatbelt is not worn. So our main aim was to detect if driver has worn a seatbelt or not while driving. - We had collected real time images from interior cameras in an automobile. These images also had various corner cases like driver wearing black clothes and had black seatbelt, or patterned clothes having black and white stripes. - Fine-tuned object detection models Mobile Net, Efficient Net for detection of seat-belt purpose. Got accuracy of 92% after building complete solution. To improve the accuracy we were working on corner-cases as mentioned in above point by using explainable AI techniques such as Grad-CAM.

Test Cases prediction using Natural language Processing Techniques.

Skills : Natural Language processing, Machine Learning, Deep Learning, Deployment

- I was working on this problem statement in my company at Bosch Global Software Technologies.
- This was a classification problem where target feature consisted of 8 components of an automobile (such as chasis, audio software, camera module,etc) - Our problem statement was to assign test cases collected from users to be automatically assigned to respective component department without manual interruption. - We used advanced deep learning like Transformer based solutions using Natural Language processing techniques as our dataset was textual. - Implemented Word2Vec model for data augmentation and used it for fine-tuning a BERT model to fetch word-embeddings. - Increased performance of existing model by 6%. Developed an API using FastAPI for production purpose.

Academic and Personal Projects :

Here are my major academic and personal projects. Soe of them are the end to end projects from data collection to model deployment on public cloud platform. The aim of these projects was to test my knowledge of ML and AI and know how to use the algorithms for various used cases. Please click on 'Github' button if you want to learn more or check the github code.

An Eye for blind - Image Captioning using Deep Learning

Skills : Computer Vision, Deep Learning (Attention mechanism), Natural Language Processing

- Built an image captioning model and read captions aloud for blind people. - Data had around 8000 images with 5 captions per image, used NLTK and Keras for captions and images pre-processing, Encoder - Attention - Decoder with CNN as encoder and RNN as decoder, transfer learning through InceptionV3 model. - BLEU score was used for evaluating the captions.
- This project is end-to-end and deployed on public cloud platform. You can try it out by clicking on button below.

Flight Fare Prediction using Machine Learning(Ensemble Model)

Skills : Data Analysis, Machine Learning

- I like to travel a lot. So for this purpose I thought it would be better if try to analyse this data and make a model out of it just for academic purpose.
- The dataset consists of records for different airlines,travel routes,stops,sources,destination,date of journeyprice,etc.Price is the target variable. - This project is end-to-end and deployed on public cloud platform. You can try it out by clicking on button below.

Sales Forecasting using Time Series Modelling

Skills : Data Analysis, Time Series, Machine Learning

- Built a model to forecast sales of a drug seller store. - Treated Outliers by capping, examined for trend and seasonality of data. - Causality tests were performed to check if other times series affect target time series or not. - Used VAR/ VARMA/ VARMAX models, exogenous variables to check performance of models, RMSE metric for evaluation.