When you are distressed about a legal matter and call a law firm for assistance, the last thing you want is to talk to an emotionless robot. CaseFlood understands that and in my role here, I enhanced the empathy and context retrieval of the AI-receptionist by applying multiple fine-tuning techniques on the Azure OpenAI Assistant API. The project currently has resulted in decrease in perplexity by 12% , reduced repitition and a warm, empathetic and efficient AI receptionist.

Automated Movie Recommender System Using Collaborative Filtering ◆Created and maintained end-to-end pipeline for an automated movie recommender system in production with event stream data (Kafka) of 1 million users and more than 30,000 movies.◆Configured model to recommend 20 movies for each request with max latency of 500ms, maintaining uptime of 99.99%◆Deployed the solution on cloud using docker, automated tests through Jenkins and monitoring in Grafana◆Conducted A/B testing and canary testing between 2 models in production and reported the differences with 95% confidence. Developed and enhanced Diffusion models for high-quality image synthesis from text descriptions ◆Incorporated QR code integration via advanced inpainting techniques.◆Utilized UNet architecture with cross-attention mechanisms to translate text into visually engaging QR code images.Developed an attention-based speech recognition system ◆Using the Transformer architecture, including the implementation of encoder-decoder modules, multi-head attention mechanisms, and strategic pre-training techniques to enhance model performanceExplainability of ML model for detecting diabetic retinopathy ◆Documented Global explanations of the model using Confusion Matrix, Precision, Recall and their interpretation for stakeholders◆Documented Local explanation for a retina scan by highlighting anchor images with 95% confidence interval using the Alibi library

◆Built ensemble models in python to detect new fraud types and reign in fraud in the system resulting in detection of 5 new fraud types and savings of $12.5 million first 8 months◆Created and maintained a central fraud analytics dashboard on Tableau, consumed by 50+ stakeholders daily

◆Led a team of 15 developers and data scientists to build an iOS app for Pest & Disease Management of Cocoa Plants with Image recognition and data collection functionalities◆The app allowed 500+ users to collect, analyze and visualize data for a 400 Hectare farm daily reducing manual effort for collecting data by 25 hours per week (per person) and increase in pest detection by ~40%

Senior Data Scientist working with a leading CPG client: ◆Oversaw a team 8 Data Engineers and Analysts to deliver an end-to-end sales reporting solution for the Middle East and Africa (15+ markets) where data was initially stored in silos across different sources◆Spearheaded the efforts to combine siloed data for Key Metric Identification, Data processing, Data aggregation, Automation of Azure pipelines, triggering of Logic apps, ingestion into SQL tables, Data model creation and visualization (Power BI)◆Estimated Impact - Reduced time for analyzing sales force data by 12 hours per week per userData Scientist for a leading Consumer Packaged Goods (CPG) client ◆Led a team of 4 data analysts on a Rationalization project of 1200+ Stock Keeping Units (SKU) across 8+ markets globally◆Created a python-based rule-engine to optimize the SKU based on sales and Supply Chain data and delivered $7.4 million in profits and paved the way for the solution to be scaled further◆Prototyped and build a dynamic reporting tool using VBA and macros on excel that was used by 120+ users

◆ Individually Handled 20+ Fortune 500 clients’ complex and ever changing needs and helpedtransform those needs into problems that can be solved by data driven decisions◆ Designed and presented proposals to clients while building a trust and rapport