Rahul Kumar

Project 2 (BGV Redeployment):

• Objective: Developed a system that leverages AWS services and AWS Bedrock's LLM to extract relevant information from uploaded PDF documents based on prompts.
• Utilized AWS services such as S3 for document storage and Lambda for serverless processing, ensuring efficient handling of user-uploaded PDFs.
• Implemented a pipeline that converts PDF documents into text format, facilitating data extraction.
• Integrated AWS Bedrock (Claude-3.5-sonnet)to analyze the extracted text and generate responses based on defined prompts, enhancing the user experience by providing relevant information quickly.
• Outcome: Enabled efficient and accurate information retrieval from PDF documents, significantly reducing manual processing time. Improved user satisfaction through the implementation of a responsive and intuitive interface.

Project 1 (Targeted Regression through AI Testing - TRAIT):

• Objective: To develop an AI based solution to optimize the identification of similar test cases within various test suites in an Investment Banking domain application, aiming to reduce time and resource expenditure during the testing phase following an application failure.
• Solution: Analyzed ~6000 test cases, each containing 5-25 test steps, to address inefficiencies for manual identification of similar test cases.
• Collected data from SWIFT files and built pipelines for further steps such as cleaning and stitching various datasets to perform EDA.
• Implemented and evaluated multiple machine learning techniques including clustering, decision trees, recurrent neural networks (RNNs) and LSTM.
• Finalized the model using decision trees due to it’s superior ability to capture and analyze the data on every sequential test steps.

Project 2 (Supplier Risk Assessment tool - LYBRA):

• Objective: Developing a Gen-AI solution that empowers LYB to proactively identify, evaluate and mitigate the risk associated with their suppliers.
• Solution: To build a framework that periodically searches external contents (News, research articles, Social Media) and provides a summarized view to category manager to make better and informed decisions.
• Worked on writing prompts using LangChain to collect data from various sources.
• Chunked and embedded the data to store in Postgres Vector DB and later retrieved using indexes.
• Used Open-AI’s gpt-3.5-turbo to summarize the chunk and provide the result to user.
• Worked on strategies to identify the best models for embedding and summarisation in the use case.

Project 1 (C3 AI Reliability Application):

• Objective: Predicting the failure of machines to reduce the downtime of plant.
• Identified the right data sources (IMS, CSI, PI in the sky) and its extraction plan.
• Helped configure Azure IoT Hub to receive and route the data to ADLS, Azure Blob Storage for organising the data.
• Performed data analytics on Azure Synapse and the data sent to C3 environment.
• Feature Engineering, Model Training and deployment was done on C3 platform.
• Used C3's native visualization capabilities to create dashboards and for monitoring the health of machines and predicting failures.

• Worked on developing Proof of Concepts (POCs) for data science projects, coupled with conducting interviews to assess candidates for similar roles

Project 3 (Liberty Global)

• Objective: Predicted the likelihood of customers to churn in order to use the insight for a proactive approach to retain the customers. Past details of ~14000 customers were collected
• Solution: To overcome the imbalance of training data EDA, Feature Engineering and Sampling techniques were applied. Classification algorithms like Decision trees, Random Forest were implemented to build models.

Project 2 (Liberty Global)

• Objective: To evaluate overall sentiment of customer towards new upgrades of software.
• Solution: Collected ~5400 Tweets and feedbacks from online discussion forum and review portal provided by consumers for text mining. Used NLP for sentiment analysis. Topic Modelling for positive and negative responses using LDA technique to ease action plan process for negative and sustain positive inputs.

Project 1 (Northern Trust Bank)

• Proficiently conducted end-to-end UI testing of web applications, Utilized Selenium to automate web browsers and streamline the testing process
• Used Pytest to organize and execute UI test cases efficiently and comprehensively
• Used Jenkins and CI/CD pipelines automate test execution and provide reporting on test results