• Designed and deployed a Retrieval-Augmented Generation (RAG)-based chatbot using the LangChain framework and GPT-4o for the legal team of system suppliers in the food, beverage, and pharmaceutical sectors, improving query response times by 50%.• Integrated citation attribution for enhanced transparency and reliability in chatbot responses.• Built robust data pipelines with SQL and Python for data ingestion, preprocessing, and dynamic index creation using Azure Cognitive Services.•… Show more • Designed and deployed a Retrieval-Augmented Generation (RAG)-based chatbot using the LangChain framework and GPT-4o for the legal team of system suppliers in the food, beverage, and pharmaceutical sectors, improving query response times by 50%.• Integrated citation attribution for enhanced transparency and reliability in chatbot responses.• Built robust data pipelines with SQL and Python for data ingestion, preprocessing, and dynamic index creation using Azure Cognitive Services.• Applied advanced chunking strategies to convert text into numeric embeddings, enhancing search efficiency by 25%.• Leveraged GPT-4 Vision, DALL-E 3, and Whisper models to implement multimodal AI functionalities, including image generation and audio transcription, enhancing user experience and content accessibility.• Conducted few-shot learning and applied prompt engineering, improving model response accuracy by 25%.• Conducted RAG evaluations to assess solution performance, ensuring high-quality deliverables aligned with business objectives.• Created Terraform scripts to provision and manage Azure Cloud infrastructure, reducing deployment time by 35% and improving scalability.• Implemented Azure Key Vault integration for secure automated secret management, enhancing security protocols and operational efficiency.• Played a pivotal role in pre-sales activities by designing POCs and presenting innovative solutions to prospective clients. Show less

• Designed and developed a binary classification model using the XGBoost algorithm to predict the risk of banking transactions, improving prediction accuracy by 20–25%, resulting in ~$5 million in annual savings.• Utilized advanced techniques like XGBoost, Random Forest, and Boruta algorithms to select the most relevant features, enhancing model precision by 15–25% and significantly reducing false alerts.• Built automated data cleansing, preprocessing, and model development pipelines… Show more • Designed and developed a binary classification model using the XGBoost algorithm to predict the risk of banking transactions, improving prediction accuracy by 20–25%, resulting in ~$5 million in annual savings.• Utilized advanced techniques like XGBoost, Random Forest, and Boruta algorithms to select the most relevant features, enhancing model precision by 15–25% and significantly reducing false alerts.• Built automated data cleansing, preprocessing, and model development pipelines using Python and SQL, reducing data preparation time by 30%.• Applied fraud enrichment techniques to detect 2–5% more fraudulent transactions previously overlooked by banking systems, bolstering security measures and minimizing financial losses.• Modeled low-frequency fraud scenarios using a knowledge-based engineering approach, reducing false negatives by 10–15%.• Conducted Population Stability Index (PSI) and sensitivity analysis to validate model consistency, ensuring <5% variation in model performance over time.• Acted as a key contributor to the design of fraud detection systems, helping NICE Actimize maintain its leadership in the financial crime prevention domain. Show less

Project II :Automated BOM Analytics:Problem: MultiLabel classification to classify the Billing of materials data into correctly configured BOM or not. Label Powerset, KNN and Naïve Bayes approach was tried on BOM data. Developed a KNN multi class classifier using R to classify the commodities into product types. Research and analysis on Sensitivity analysis, Cardinality, Label density and class imbalance. Multiclass classification to… Show more Project II :Automated BOM Analytics:Problem: MultiLabel classification to classify the Billing of materials data into correctly configured BOM or not. Label Powerset, KNN and Naïve Bayes approach was tried on BOM data. Developed a KNN multi class classifier using R to classify the commodities into product types. Research and analysis on Sensitivity analysis, Cardinality, Label density and class imbalance. Multiclass classification to classify the Billing of materials data into correctly configured BOM or not............................................................................................................................................................................Project I: Predict the cost of the shipment [Freight Management]Problem: Given a request from a customer for shipment, find the estimated cost for the Carriers. This will help to quote the best price for the customer and reduce the revenue leakages and increase the bottom line. Built a cost prediction model using Bagging ensemble method.Bagging ensemble was implemented on combination of Linear Regression, Ridge regression, Lasso regression, elastic net.Tableau is used for data visualization.Variance inflation factor was used to detect the multicollinearity between the variablesPrincipal component analysis (PCA)was used to identify the features which explains more variance in the target variable. Show less

Part of Platform support team(L3) in Mortgage Technology.