Swanand Mhalagi

Stock Data technical Analysis*

Data Redesigning\Visualization, Web Scraping, Pandas, Plotly

- Developed a tool to analyze and visualize some of the key technical indicators.

- Built a single function to plot candlestick, moving average and MACD patters. Wrote a web scraper script to visualize analyst recommendations from multiple websites on a single page.

Coronavirus Data Analysis with predictive models

Plotly, Prophet, Time Series, Python

- Build a region-specific data analysis dashboard for Coronavirus with predictive capability for flattening of the curve.

Recommendation System with Python*

Sklearn, Pandas

- Evaluating recommendation system algorithms such as user and item-based clustering, collaborative filtering as well as hybrid recommendation approach for discovering potential preferences of individual users.

Conquer Class Imbalanced Dataset Issues using GAN

GAN, TensorFlow, Image Generation

- Used Deep Convolutional Generative Adversarial Network (DCGAN) to generate images of certain class types to improve image classification on imbalanced datasets.

- Implemented several tricks for generator and discriminator from research papers to make GAN stable.

The Quest of Higher Accuracy for CNN Models

CNN Accuracy, Performamce, Image Classification

- Implemented performance improvement techniques such as hyper-parameter tuning, data redesigning and model optimization.

- Defined easy to use data processing pipeline. Laid-out a methodology to use optimum hardware resources. Addressed imbalanced data base issues. Improved accuracy from 68% to 82%.

CNN Implementation for Diabetic Retinopathy Detection

CNN, Image Classification, TensorFlow

- Developed a convolutional image classification system for diabetic retinopathy detection.

- Defined alternate methods of implementation. e.g. loading images with OpenCV as well as NumPy array.

Graduate School Recommender using Spark ML

Spark ML, Hadoop

- Designed machine learning application to streamline the graduate school application processes.

- Tool recommended best suitable schools for applicants according to their academic profile such as GPA, GRE score and extracurricular activity.

Software Defined Storage

CEPH, OpenStack Swift and Cinder, Cloud Performance, Benchmarking

- Built CEPH SDS cluster with CMR and SMR drives for big data applications.

- Characterized performance with Iozone, Fio and DD benchmarks. Deployed and evaluated OpenStack Swift performance with Intel COS benchmark.

Cloud e-lab for OpenStack and Deep Learning

Python, Private Cloud, Cloud Application

- Developed backend system to create pre-installed environments for deep learning and cloud developments with just one click.

- Wrote interactive tutorials for deep learning frameworks and OpenStack services.

Galaxy Star Count

OpenCV, OpenMPI

- Python code to count number of stars in a galaxy image released by NASA

- Used collective communication to improve and parallelize the count processes.

Cloud Software Engineer

Intel Corporation

Cloud Solution Fabric: Accountable ownership of project from concept to production. A containerized tool allows single click Azure Stack HCI deployment. Developed unattended Windows deployment solution.

AI Model Performance: Laid out efficient container deployment methodology on CPU and VPU for AI inferencing on Azure Stack Edge platforms with various precision models. Developed and optimized deep learning TensorFlow models for Resnet50\SSD_VGG16\BERT to analyze performance of INT8 instructions used in AI inferencing.

Workload Development: Contributed to several Intel proprietary benchmarks development to cater engineering needs. Provided data driven recommendations to Microsoft Azure\AzureStack teams for software stack and platform configurations.

Feature Enabling: Demonstrated Intel products value proposition to OEMs/CSPs. Enabled Intel products like Fortpark, Optane and DC PMM for Microsoft cloud stack.

OpenStack Developer Intern

UTSA and Intel Corporation Colaboration

- Acquired expertise in deploying OpenStack services such as Nova, Keystone, Neutron, Swift, Cinder, etc.

- Developed Network automation scripts for Neutron in python.

- Developed prepackaged scalable cloud instances for deep learning application development on top of GPU and CPU.

Graduate Research Assistant

Open Cloud Institute, UTSA

- Deployed Open Compute (FB) servers for OpenStack and Machine Learning purposes.

- Built CEPH and Swift distributed software defined storage cluster for big data applications and analyzed performance with various storage benchmarks.

- Developed Ansible scripts to deploy web apps and to update Linux packages on multiple servers simultaneously.