Improved BrewRight, ABI’s legal analytics platform for detecting fraudulent transactions for financial compliance

• Developed machine learning models and implemented hypotheses to flag anomalous and risky transactions
• Set-up CICD for DevOps pipelines, optimized API data extraction scripts and achieved E2E Automation
• Migrated existing codebase to run in PySpark using Databricks; reducing runtime by 70%, saving USD 100,000

NDA Robot Design

• Responsible for the design of a low-cost, modular surgical system to guide invasive surgical operations.
• Designed to facilitate 360 degree access to the needle guide during surgical operations; Controlled through servo motors.

Worked on energy and environment management, focusing on transport of toilet soap production materials

• Achieved over 30% reduction in line losses in the form of condensate, saving over 10 million INR/year
• Introduced automized steam trap-monitoring wirelessly and proposed a two-stage desuperheater system for precise temperature control at the machine end, reducing energy wastage by 20% and estimated cost reduction by 10%
• Designed the piping layout of the Hyderabad factory of Wipro, capable of producing 20L soaps per day

Worked on predicting the work in progress (WIP) stock of the assembly shop.

• Optimized the process by making it data driven and helped reduce the WIP stock by 9%
• Reduced monthly production shift planning time by 80%

Showcased video cameras as a quick and inexpensive way for remote, non-contact vibration analysis

• Scale invariant features and optical flow algorithms used to extract the motion signal from the video frames
• The peaks obtained from FFT plots generated using the motion signal were found to be in 8% of the body’s NF

AI based hackathon to diagnose pneumonia through abscesses or pleural effusions in X-Ray images

• Tackled as an object detection problem, with only one type of image to detect. Yolo-v3 from Darknet was used
• Achieved an IoU of 46.97% and MAP value of 35.44%.

Developed an ML model for prediction of Indian rainfall on monthly and seasonal time scales

• Used RNNs to forecast rainfall for June to Sept (ISMR) by feeding data on rainfall and sea surface temps
• Set-up CICD for DevOps pipelines, optimized API data extraction scripts and achieved E2E Automation
• Temperature was used to augment, accounting for the fluctuations; the model achieved an RMS Error of 24 cm

Specifics -

• Created a machine learning based model to predict purity of the Iron ore (percentage of Silica Impurity) prior to froth floatation, to enable engineers to take corrective actions and ensure that the product quality remains high
• The model employed random forest classifiers, neural networks and XGBoost in an ensemble to predict the output Silica concentrate based on input ore properties, ore pulp condition, air flow rate & other process parameters
• The predicted value of silica concentrate in the ore was found have an R2 score of 0.941 in the testing set

2-tier Hackathon on sustainable development, organized by CII; contested by 300+ teams across 7 countries

• Ideated, pitched, & showcased Dairy Folk: A start-up focusing on improving livestock management using unique IDs
• Enabled continuous health monitoring of livestock by decentralizing storage of data for enhanced transparency

Specifics -

• Created a web crawler using selenium & bs4 for maintaining offline copy of course documents from Moodle.
• Wrote a daemon to refresh the offline version daily; File system updated to match changes in course structure

This was implemented using Arduino based PID control utilizing data from a gyroscope &. An accelerometer in a closed feedback loop to control the rotation speed of a flywheel, thereby maintaining the unstable equilibrium.