Voice recognition systems facilitate the interaction with technology, enabling hands-free requests. With that thought in mind, This project was to create a working model voice controlled robot arm i.e. an arm that is capable of motion using only voice inputs. The system is capable of recognizing a trigger word and then wait for a voice command and move accordingly. The two models are built using recurrent neural networks.

This project was done as a Team Project, with me acting as the team leader. It involved mining data from twitter and carrying out sentiment analysis on the data. The Natural Language Processing model was trained of a separate dataset and used to predict the sentiments. Visualization was done on tableau and can be found here

The project goal is to train machine learning methods to recognize different pathogen objects, and to make this accessible in the form of an Android application usable at the point of care. This work began with machine learning methods based on extracting statistical characterizations of the shapes in each image. Deep learning techniques involved data augmentation for a robust model, early stopping and model checkpoint to avoid overfitting.

This project involved creating a robust predictive model that would help identify identify people who would subscribe to a bank term deposit. it involved carrying out data exploration, data cleaning, feature extraction and creating a model that can run from a command line. Techniques such as dimension reductionality using PCA and T-SNE and fixing class imbalances using SMOTE were employed in the project

This project was entered for the Nairobi Tech Week hackathon in 2019. The project involved creating a Convolutional Neural Network for crop disease detection, data augmentation for the small image dataset, and working under pressure as the whole project was built in two days.

To Improve my skill set, I decided to learn Golang using the book by Alan A.A Donovan and Brian W. Kernighan called the The Go Programming Language. The book involved concepts such as Goroutines and Channels, Concurrency with Shared Variables and Reflections

Using asynchronous programming, I was able to fetch data from the usgs-lidar-public that contains cloud point data for various regions in the united states. The challenge was to extract elevation from the cloud point data to improve the understanding of how water flows through a field, this in turn would help better the understanding of which areas will produce more when the same fertilizers, machinery is applied to the whole field