SURYA SAMARTH J
Work Experience
Drexel University
Project - Acoustic Scene Classification of Heart Sounds
Dataset Analysis and Preprocessing
-
Analyzed the Circor Heart Sound dataset from PhysioNet.
-
Reduced noise and extracted crucial features such as MFCCs.
-
Computed STFT, wavelet transform coefficients, and EMD coefficients using OpenSmile.
Deep Learning Implementation
-
Utilized CNNs, LSTMs, and SVM for heart sound classification.
-
Classified heart sounds into normal or abnormal categories based on the presence of murmurs.
-
Applied PCA and t-SNE for dimensionality reduction to simplify the data while preserving essential patterns.
Automated Cross-Validation
-
Implemented automated 5-fold cross-validation for robust data analysis.
-
Utilized extracted features for model evaluation.
-
Demonstrated expertise in preprocessing, feature extraction, and performance analysis to ensure reliable model performance.
Data Science Research Assistant
September 2023 - December 2023
Anmerkung Solutions Private Limited
Data Scientist
March 2022 - August 2022
AI/ML In-House Tool Development
Played a crucial role in developing an in-house 2D annotation tool at Anmerkung Solutions to semi-automate the annotation of vision-based automotive data.
-
R&D Initiatives: Contributed to fine-tuning MobileNet and YOLO models, optimizing them for efficiency and accuracy in real-time object detection, essential for autonomous driving applications.
-
Backend Development: Worked within a 6-member team on backend development using Django and Python, focusing on data handling and API integration to ensure swift and accurate processing of large automotive datasets.
Sign Board Detection
Involved in a project aimed at developing a custom YOLO-based model for detecting road signs, specifically tailored for Indian road conditions.
-
Model Development: Assisted in the development and optimization of a model trained on 3,000 images, covering 8-10 signboard classes. Achieved an 85% accuracy rate by focusing on true positives and minimizing false negatives, reducing the need for manual corrections.
GD&T Symbol Recognition
Participated in a project to develop an OCR tool for detecting Geometric Dimensioning and Tolerancing (GD&T) symbols in 2D mechanical drawings.
-
OCR Tool Development:Collaborated with a team to develop a tool that achieved 90% accuracy in detecting GD&T symbols, streamlining the quality control process in manufacturing by reducing manual inspections.
Weezy Streetwear
Content Creation and Marketing
-
Joined as a freelancer, curating, shooting, and editing video advertisements.
-
Contributed to the brand's marketing campaigns and enhanced its online presence.
Digital Content and Marketing Strategies
-
Delivered 8 professionally shot and edited video advertisements, leading to a significant increase in online engagement over six months.
Supply- Chain Analysis
-
Reduced overstock by 30% through trend analysis and inventory optimization, enhancing inventory management.
-
Developed a dynamic supply chain model that balanced demand and supply, resulting in an inventory turnover of 30 days.
-
Streamlined data processing and insights generation by designing and maintaining ETL pipelines, ensuring clean and reliable data for inventory and sales analysis
Funnel optimization [Product Listing Page(PLP) to Product Detail Page(PDP)]
-
Boosted add-to-cart rates by 1.2% and improved final conversion rates by 0.3% by identifying key drop-off points and implementing targeted discount strategies
Cross-Functional Collaboration
-
Acted as a liaison between marketing, product development, and operations teams.
-
Aligned data-driven strategies with business objectives to optimize processes and boost profitability.
Sales and Demand Forecasting -
Trend Analyst / Content Creator
July 2021 - April 2022
Anmerkung Solutions Private Limited
Power BI Dashboard Creation
-
Quality Control Feedback Loop: Developed a Power BI dashboard to monitor annotation quality, using a feedback loop to highlight rejection reasons and error types, thereby improving annotator performance.
Object Detection Domain
-
Trained on an OEM proprietary annotation tool, applied AL/ML techniques to derive ground truth data, improving model accuracy and boosting the KPI index by 2%.
Image Labeling and Segmentation
-
Annotated 300 images with CVAT using bounding box and segmentation techniques, and 12,000 images with an OEM proprietary tool, increasing the daily annotation rate.
High Beam Assist
-
Derived the blockage percentage and visibility threshold key point using ground truth data, reducing the algorithm’s delay in response.
(Internship)
Data visualization and Analysis intern
March 2021 - August 2021
Curious about my professional escapades? LinkedIn is your next stop
My LinkedIn
(click on the logo)
Also, here is my fancy sheet of paper
My Resume
(click on the logo)