Naresh Kumar Karanam

• 1 . Lost and stolen parcels prediction at last mile delivery Developed a xg boost classification model to predict the lost and stolen parcel at last mile delivery and successfully deployed the model in AWS environment
• 2. Predictive Modelling for Delivery Time Window Estimation during Round Start | Client: Belgium Post EU
• - Developed probabilistic model using Random Forest, XGBoost Classification, predicting precise slots (< 3 hours), reducing uncertainty by 50%
• - Incorporated a refinement model to feed on ground-real time insights like delay & delivery sequence change to refine predictions dynamically
• - Implemented an adapter detection to identify positions following delivery sequence, adopted transfer features to integrate vehicle patterns
• - Conducted exploratory data analysis using Pandas, Numpy, visualized distribution using plotly, matplotlib, plotted geo-spatial data using folium
• - Used KMeans Clustering for dimensionality reduction, implemented Explainable AI (XAI) using SHAP to enhance model interpretability
• - Implemented multi-layered model including inference, optimization & business logic layer & added guardrails to prevent wrong predictions
• - Improved prediction accuracy to 88%, with 70% parcels having 1-hour window, enhancing customer satisfaction and logistics planning
• 3. Design Gen AI QA Pipeline for HR Question-Answering App Backend | Client: PostNord EU
• - Engineered LLM-powered AI Q&A system using Retriever-Augmented Generation, Gen AI & NLP, reducing query resolution time by 50%
• - Created vector embedding using AWS Bedrock Titan, OpenAI text, stored in vector databases like Chroma, Pinecone for KNN similarity search
• - Implemented Retriever Chain by combining Retrievers and prompt engineering to ensure efficient retrieval, used vector indexing like HNSW, IVF
• - Applied Map-Reduce, K-means for text summarization using Anthropic Claude LLM, used prompt engineering & guardrails for fine-tuning response
• - Deployed backend inference pipeline as Docker container in AWS ECS, setup ECS to run image using Fargate, ECS, optimizing infrastructure cost
• 4. Time-Series Forecasting of Parcel Volume Expected to be Distributed over Next 4 Weeks | Client: Belgium Post EU
• - Developed forecasting model using ARIMA, SARIMAX, GARCH to predict parcel volume, considering workforce migration & associated costs
• - Analyzed time-series to isolate long term effect, detected multiple seasonality, used auto-correlation & moving average to model the data pattern
• - Conducted root cause analysis on fluctuations, flagged anomalies, events using categorical features, used Lasso to explain model behavior
• - Increased forecasting accuracy to 90%, reducing delivery costs by 6%, and deployed results for end-user consumption via AWS Lambda & Power BI
• 4. Statistical Analysis of Employee Well-Being Survey to Identify Impacting Factors on Sentiment | Client: PostNord EU
• - Conducted multivariate analysis to determine key factors affecting employee well-being and estimate sentiment at different organizational levels
• - Checked bivariate correlation strength between well-being index and individual factors using Pearson, Spearman. Used Statsmodel OLS for multivariate regression & hypothesis testing, used multiple correlation, coefficient of determination to analyze component effect to mitigate issues
• - With strong presentation, conveyed trend analysis to non-technical leadership to guide strategic decision making, boosting productivity by 20%
• Role: MLOps Engineer
• - Streamlined MLOps pipelines with CI/CD, Git, Bitbucket for deployment using AWS Sagemaker Endpoint, ECR, ECS, reducing failures by 50%
• - Developed scalable real-time inference pipeline using AWS Kinesis Enhanced fan-out, Sagemaker Multi-model Endpoint with auto-scaling
• - Deployed batch inference pipeline as Docker container in AWS ECS, setup ECS to run image using Fargate, ECS, optimizing infrastructure cost
• - Orchestrated automated retrain & prod deployment using CI/CD, Git, Bitbucket, Cloudformation, CodeBuild, CodePipeline to ensure stability
• - Integrated Sagemaker training job & Hyperparameter tuning for model retraining & A/B testing using multi-variant endpoints in prod ML workflow
• - Used MLflow for experiment tracking & deployment, developed custom model evaluation KPI for monitoring accuracy, data drift & data integrity.
• - Configured monitoring using AWS Cloudwatch Logs, setup Alarm to send email alert on failure using SNS topic for enabling quick resolution
• - Integrated SonarQube based quality checks for unit testing to ensure proper code coverage & reliability, also used mocking frameworks.
• Led cross-functional teams to develop predictive models improving decision-making processes.
• Mentored junior data scientists, fostering professional growth and knowledge sharing.
• Collaborated with stakeholders to identify business requirements and translate them into data-driven solutions.
• Developed data visualization tools that improved reporting capabilities for executive leadership.

.

• Analyzed large datasets using Python and SQL to identify trends and insights for stakeholders.
• Implemented machine learning algorithms that improved accuracy of forecasts and predictions.
• Delivered comprehensive reports and visualizations to communicate findings to non-technical audiences.
• Streamlined data collection methods to minimize analysis errors.
• Conducted feature engineering efforts to enhance model performance by creating new relevant variables from raw input data sources.

• Developed and optimized PLC programs for automated machinery operations.
• Conducted troubleshooting and maintenance on PLC systems to ensure operational efficiency.
• Collaborated with engineering teams to integrate new technologies into existing systems.
• Designed user-friendly interfaces for operators, enhancing usability and reducing errors.
• Implemented process improvements that increased system reliability and reduced downtime.
• Analyzed system performance data to identify areas for optimization and enhancement.
• Reduced downtime by troubleshooting and resolving complex issues with PLC-controlled equipment.