Anish Joshi

Health Services Advisory Client

• Classified physician notes by building document classification model using Logistic Regression to segregate substantive text records. Further classified them into broad classes using Multinomial Naive Bayes algorithm. Utilized pySpark to distribute data processing in the pipeline which correctly mapped 80% of records to existing structured categories.
• Transformed healthcare data as per HIPAA guidelines by utilizing Keras LSTM-RNN to generate artificial data to replace existing training data used to supplement NLP tasks.
• Constructed Named Entity Recognition model and deployed it in production-grade environment using Docker containers and Aqua 2.5. Monitored model performance and maintained 92% accuracy in discovering relationships among biomedical entities.
• Manipulated complex oracle databases using SQL as part of data warehouse management and ETL pipeline processes to develop of custom Tableau dashboards to visualize patient safety events in hospitals.
• Facilitated exploratory data analysis and conducted statistical tests in SAS and R to evaluate model performances.

Client: Visa

• Performed Time-Series hierarchical clustering and bootstrapping to extract value from large time-series dataset using SciKit Learn and Scipy.
• Used Principal Component Analysis to improve robustness of predictive model by 30%. Synthesized findings and presented recommendations using Tableau and engaged with leadership to develop bespoke solutions, collecting relevant data and creating high-quality deliverables.

Data Warehouse Restructure

• Reconstructed entire data mart. Enhanced performance by 43% (data obtained from Solar Winds).
• Scripted Replication to transfer ~2 billion records per second.
• Performed A/B testing (Dev,Test, Prelive and Beta environments to production/ Live environment).
• Managed team of 5 interns to create homegrown API - similar to Kafka - entirely written in ASP.Net.
• API was able to handle 1 million records per second.

Object Detection - OpenCV

• Created a tool which can start job as soon as car enters bay and end the job service automatically when car leaves bay
• Tools Used: Jupyter and AWS for Production
• Custom Object detection – Created through labelImg and YOLO
• Huge financial savings for company. Initially company was using RFID technology and third party tool – costing ~$100,000 annually
• Revised tool brought down cost to ~$10,000. Cost spent only on maintaining new camera devices and tool

Behavioral Analysis

• Developed script using python for Test and Training ~2TB of data (Used 80-20 partitioning scheme)
• Used XGBoost and Tensorflow for post-production
• Used AWS Redshift for deployment on AWS Neptune
• Tools used: AWS EMR and Spark for data pipelining
• Developed intricate algorithms based on deep-dive statistical analysis and predictive data modeling.

Customer Risk Analysis

• Applied Predictive analytics to determine if customer prone to fraud in future
• Statistical tool: Stata, R
• Used TensorRT, Keras, Tensorflow, PyTorch, CNN and RNN
• Used AWS Redshift for deployment on AWS cloud
• Algorithm used - Regression Classification
• Accuracy achieved - 92%