Narender Kumar

Retrieval Augmented Extraction using Fine-Tuned LLaMA 3.0

• Addressed the challenge of hidden insurance coverage buried in documents by implementing a RAG-based language model, surfacing 25% more recovery signals beyond structured data.
• Converted text to embeddings using Word2Vec, built a semantic retrieval layer with chunking using LangChain and cosine similarity to identify top-k relevant chunks per query.
• Fine-tuned LLaMA 3.0 using LoRA adapters to extract structured coverage fields from retrieved context, significantly boosting domain-level precision and consistency.
• Validated model by comparing the similarity between desired and actual embedding output and expert-reviewed feedback, deployed via FlaskAPI with Azure Container Apps, resulting in a 10% increase in recovery success and 18% faster claim cycle time.

AI Agent Developer
Tools: Python, Agno, Claude/GPT-4o, YFinance, DuckDuckGo, Vector Embeddings, Generative AI

• Designed and deployed autonomous AI agents using Agno, achieving faster startup times and 50x lower memory usage compared to conventional agent frameworks.
• Developed multi-agent systems for financial reporting, real-time web search, and knowledge retrieval leveraging vector databases.
• Integrated APIs such as YFinance and DuckDuckGo for dynamic stock analysis, real-time fact-checking, and source attribution.
• Enabled multimodal support (text, image, audio, video) and implemented long-term memory capabilities to enhance agent context retention.
• Delivered structured markdown outputs and built real-time agent monitoring dashboards to improve transparency and performance tracking.

Collection Models (Predictive Modeling and Deployment with Azure)-As part of an internal tool, developed machine learning models to predict customer personas like high risk and low risk for unpaid bills. Subsequently, used MLOps (Azure DevOps and Azure services) to create a tool that enables clients to upload their customer data regularly and receive individualized predictions. These predictions empower clients to employ services like automated calls and messages for effective reminders about unpaid bills.

Credit Card Bills and Internet Service Bills Collection Optimization–

• Developed cutting-edge recommendation engine for forecasting customer issues, resulting in enhanced issue resolution and improved customer satisfaction.
• Built predictive model for credit card payment propensity, contributing to a 25% reduction in late payments and 15% increase in on-time payments.
• Transformed processing pipeline to Azure Function Apps and Container Instances, achieving impressive 80% reduction in processing time.
• Streamlined prediction process by migrating preprocessing and prediction to server, significantly enhancing overall efficiency and performance.

SKILLS – Machine Learning, Statistics, Azure DevOps, Azure, Docker, Flask

Interaction Insights of customer (NLP using generative ai(ChatGpt) and Azure)

• Created a tool capable of extracting insights from text data in six different languages. specifically focusing on chat conversations between customer service agents and customers.
• This tool categorized text data into four key categories - Agent, Process, Service, and Technology - and predicted customer loyalty, churn, journey stage, effort, and intent behind customer contacts using ChatGPT API for advanced language processing.

SKILLS – NLP(Natural Language Processing), Azure, OpenAI, Mysql

Scalable Data Pipelines for Business Intelligence and Advanced Analytics

• Designed and implemented robust data pipelines integrating APIs, SQL/NoSQL databases, and flat files into Azure Synapse Analytics, accelerating data flow for ML models by 30%.
• Standardized data transformation processes using Azure Data Factory (ADF) and Python, ensuring seamless transitions across staging, warehouse, and semantic layers for enriched reporting.
• Automated workflows with Azure Data Factory pipelines and Azure Data Lake integrations, eliminating 90% of manual dependencies and delivering timely, high-quality data to Power BI dashboards.

Wafer Fault Detection - Machine Learning Project

Developed a Wafer Fault Detection system using Machine Learning to classify wafer sensors as faulty (-1) or functional (+1) based on sensor data.

Key Responsibilities:

• Data Validation: Automated validation of sensor data files based on schema rules (file name, column count, data types, missing values).
• Database Management: Created and managed a structured database to store validated data for model training and prediction.
• Feature Engineering: Imputed missing values using KNN imputer, removed redundant features, and applied KMeans clustering to segment data.
• Model Training & Selection: Trained Random Forest and XGBoost models for each cluster and selected the best-performing model using AUC score.
• Prediction Pipeline: Integrated pre-trained models into a Flask API for real-time wafer quality classification.

Technologies Used: Python, Pandas, NumPy, Scikit-Learn, SQL, Flask, Random Forest, XGBoost.