Harshith Sakala Santhosh

Graduate Machine Learning Engineer A recent graduate in applied data science with 3 years of experience in building Machine learning and Deep learning projects. Have proficiency in Data Preprocressing, Model building, as well as scripting languages, Python, Java and SQL. Awarded the best student of the year in the Engineering Department.

Capuchin Bird Deep Audio Detection (02/2023 - 03/2023)

• Constructed a deep learning model from scratch using TensorFlow, Sklearn that identifiedthe number of capuchin birds in a single area from audio clips with 92% accuracy
• Project Link: https://github.com/sakalaharshith/Capuchin-Bird-Deep-Audio-Detection
• Technologies used: CNN, Tensorflow,SKlearn, Pandas, MatPlotlib, Python

Poem Generation through LSTMS (10/2022 - 10/2022)

• Developed a poetry generation project utilizing LSTM's and TensorFlow to train a modelagainst Shakespeare's sonnet data, achieving an accuracy of 98%; the project was able togenerate poems based on a small input sentence
• Project Link: https://github.com/sakalaharshith/Poem-Generation-through-LSTM-s
• Technologies used: LSTM, TensorFlow, Matplotlib, Pandas, Python, Google Colabs

Leaf Disease Detection and Segmentation through Transfer Learning(07/2022 - 09/2022)

• Spearheaded a leaf disease detection using cutting-edge ML models such as Mask
• RCNN(Instance Segmentation), Efficient Net B1, and Inception Net V3(Image Classification)and trained against the Plant Village Dataset, the project successfully achieved a 71% mAPwith Mask RCNN and > 95% accuracy with Efficient Net and Inception Net
• Implemented a state-of-the-art Mask RCNN for identifying leaf disease; surpassed theperformance of traditional image classification models (Inception Net, Efficient Net) by 40%when tested against plant village dataset of with no background images and field imagestaken from internet
• Project Link:https://github.com/sakalaharshith/Leaf_Disease_Detection_instance_Segmentation(Leaf
• Disease Segmentation)
• Project Link:https://github.com/sakalaharshith/Leaf_Disease_Detection_Image_Classification.git(Leaf
• Disease Classification)
• Technologies used: Mask RCNN, Efficient Net B1, Inception Net V3, TensorFlow, Matplotlib,
• Sklearn, Python, Google Colabs

Recognition of human activity from smartphone sensor (03/2022 - 04/2022)

• The project aims at recognising of human activity from smartphone sensor readings.
• Data cleaning, Data Preprocessing like Data balancing(SMOTE) , feature selection (Mutualinfromation), feature extraction(PCA) was performed and trained using SVM,KNN, Decision-trees classifiers
• Decision-Trees trained from feature selection (Mutual information ) data outperforms theother models with an overall accuracy of 98.56% and atleast of 95% accuracy in each model
• Project Link:https://github.com/sakalaharshith/recognition-of-human-activity-from-smartphone-sensor-signals-
• Technologies used: SVM, KNN, Decision Trees, PCA, Mutual Information Classifier, SMOTE,
• SKLearn, Python, Google Colabs.

Hospital Cypher-A Smart Surveillance Approach (03/2021 - 06/2021)

• Collaborated with a team of 5 to design and implement Hospital Cypher, a Python-basedserver-desktop app that reduced the spread of COVID-19 infection by 20-25%; achieved a98.35% accuracy rate using MIT's Deep Learning model for face recognition technology
• In addition to that, The team also developed a face mask detection model using SVMalgorithm that accurately identified people without face coverings with 93% accuracy,enabling hospital authorities to take necessary action in real-time
• Project Link: https://github.com/sakalaharshith/Hospital-CypherA-Smart-Surveillance.git
• Technologies Used: Face Recognition model by MIT, SVC (SVM), Pandas, OpenCV, Python,Visual Studio Code.

Deep Learning Foundations: Natural Language Processing with TensorFlow

A Research Journal on Hospital Cypher- A Smart Surveillance Approach in IJCRT (05/2021 - 06/2021) 

• This research paper journal is based on hospital cypher which is a bachelors dissertation project. The link to Research paper: https://ijcrt.org/papers/IJCRT2104588.pdf