MLOPS | Demand Forecasting | Natural Language Processing(GEN AI) - Developed MLCore, an end-to-end MLOps platform for healthcare, offering real-time updates from data ingestion to model deployment. - Features customizable EDA, testing, and monitoring modules. - Added the feature for Model fairness and Bias evaluation for the models logged in the application( MLCore). - Built the solution for HR job description generator using RAG, GPT3.5 and optimized the token count and dollar costs by prompt engineering techniques. Decreased the latency by 60%.- Reduced cost and memory usage by using various optimizing strategies in the vector database (Milvus).-Built POC on generating the SQL codes using LLM and summarizing & generating the business insights from user prompts in natural language. Followed the best practices to address Data Privacy and Data Security. Worked closely with Devops team to migrate this APP from Streamlit to the web application built on top of Django framework. - Led the development of an end-to-end inference pipeline for Demand Forecasting on Azure Data Factory. Adhered to industry best practices to create a versatile template capable of executing various forecasting use cases within the pipeline. • Incorporated a new capability enabling data ingestion from cloud services (AWS, GCP, Azure) into our internally developed EDA and model evaluation dashboard.

DEMAND FORECASTING Integrated the ML models with WEB-APP to build the scenario planner using Django framework. • Worked closely with Data Scientists to seamlessly transition from the inference stage to the scenario planner stage, facilitating informed business decisions. • Achieved a notable 40% speed enhancement by optimizing queries and refining the database schema.

Project: Usage Billing SystemTech Stack: Spring, Angular, Typescript, MySQLDescription: Usage Billing System is a web application where it prints the real time data usage of the clients and shows the total amount to be paid at the end of the month as per the service consumption. The real time data usage is generated with the help of simulator that generates the IPDR records (contains IP add, MAC, Bytes sent/received, Hostname) for each user session. The billing system needs to process the usage in each IPDR records and sums it up for each client.

Learnt many new things about Ongoing inside the Software Development in an #Investment_Banking Firm.Program Deliverables:->Interface with a stock price data feed->Use JPMorgan Chase frameworks and tools->Display data visually for traders(Trader's Dashboard)->Bonus task: Open source contribution