Conducted research on Large Language Models (LLMs) to mitigate diffused modal collapse in AI models by exploring strategies like Reinforcement Learning with Human Feedback (RLHF). Utilized HelpSteer dataset to prevent self-training on generated text, ensuring adaptability. Augmented model performance through ensemble learning for enhanced generalization across diverse datasets.

Engineered a web project using HTML, Bootstrap, and Tailwind CSS to optimize interactions with nearby hospitals. Integrated Google Maps API with JavaScript and AJAX to enhance user engagement in a single page. Optimized search time latency and reduced server requests by 30%. Integrated text fields with auto location suggestions using Google APIs to direct users' searches to specific hospitals with zero errors.

Link

Collaborated with an innovative startup to reduce computational costs by creating a serverless AWS web application to upload local folders into S3 buckets. Configured OIDC authorization with API gateway to DynamoDB for manipulating text log files.

Implemented YOLO AI model using segment detection algorithms and calibrated error distancing using time series Dynamic Time Warping (DTW) to achieve real-time human pose detection. Achieved an impressive 89% detection accuracy from UCF101 over 600+ videos.

Engineered a real time android application using Java that makes use of camera and depth sensors to permit disabled people to interact and learn over 200 sign languages.

Enhanced the functionality by using PoseNet which was used to categorize multi set data points for detecting more than one human in a single frame.

Fabricated a GUI based application utilizing Java, JavaFX, Swings, SQL that uses an interface and a parser to fetch sensor values from CAN Bus dataset for autonomous vehicle driving

Created a real-time Advanced Driver Assistant System predictive classifier to decode vehicle sensor values to forecast future curves by 33% ahead of time matching the exact GPS Coordinates and performed analysis on the frequency of distribution for all segments.

Link

Developed an ecommerce platform using Full Stack, MongoDB with Node JS by engaging with multiple customers and designed process matrix to scale up the business. Extended the capabilities which can used to dynamically change the space for advertisements as and when needed

Optimized with best measures with a page TBT response time less than 70ms and enhanced the productivity by incorporating incident management system with various priority levels (P1, P2, P3) which extends support to clients.

Engineered a centralized admin monitoring system, which can be used to observe daily profits and dynamically change the deigns. Added additional capabilities for delivery executives which can be served as in real time for delivering orders to the customers.

Link

Curated an ML model in an attempt to strategize an advertisement targeting based algorithm based on age groups.

Created an android app which tracks Human Touch Gesture patterns. Conducted research and compared the analysis with various other models like Random Forest and SVM, which speculated more usage of a Fling gestures during ads.

Designed an automated workflow to assemble all submissions by students using bash scripting. Verified each submission with plagiarism checker by automating and creating mapping to MS Excel.

Integrated OneDrive in VM and created folders for each student thereby notifying when their submission is processing.

Devised a GUI based hospital management application by using Java, JavaFX and MySQL. Enhanced the application by adding a chat functionality to send priority messages to doctors upon scheduling appointment.

Designed a game for creating hurdles in a maze and created an algorithm to travel through starting and ending points in C language. Enhanced the game by adding various hurdle levels and calibrated the accuracy by joining both ends.