Responsible AI
Responsible AI is the foundation for creating AI-based solutions that adhere to ethical guidelines and societal values. It serves as a guiding principle, ensuring that artificial intelligence systems are developed and deployed transparently, fairly, and accountably. It involves establishing governance frameworks that foster trust, enabling stakeholders—including policymakers, researchers, and industry leaders—to collaborate effectively. Through this approach, Responsible AI seeks to harness the benefits of AI while minimizing risks, Check out latest AI Risk Management Framework developed by NIST.
AI solution portfolio
This collection of AI-based applications and projects by Nehal Naik showcases the various components and tools, which can be used to develop enterprise-level AI solutions. These projects address diverse use cases across multiple industries, demonstrating the potential and versatility of advanced AI technologies.
Retrieval-Augmented Generation (RAG) AI framework bridges the gap between
LLMs that power generative AI and private or proprietary data sources. This grounding technique helps build robust knowledge base and reduces hallucination.
​This RAG based customer support chatbot developed using GCP Agent, retrieves information from company’s private knowledge bases (Personal Resume and Camera Manual is used as an example) and generates accurate responses to customer queries, improving satisfaction and reducing support workload.
Sentiment analysis utilizes Large Language Models (LLMs) like GPT to classify text sentiments (positive, negative, neutral) by analyzing language patterns and context. ​​
This AI based Customer support agent bot leverages GPT 3.5 model and few shots prompt which can assess user comments, enabling automated responses that match the sentiment detected (positive, negative, neutral). This method streamlines customer engagement, ensuring efficient handling of social interactions while enhancing customer satisfaction through personalized responses.
This project utilizes historical auto insurance data and claim filing patterns to predict the likelihood of claim submission based on various policy parameters. The data underwent a thorough cleanup and engineering process to ensure its quality before model training. A Random Forest model was employed for training, and classification metrics were used for evaluation.
The model is currently deployed on Google Cloud via a Docker image, serving requests through an API. A sample application has been developed to demonstrate the model's core functionality. This setup is designed to be easily scalable, allowing for the management of larger volumes of policies within a flexible GCP architecture.
This project leverages structured hotel booking data within RapidMiner to develop and test machine learning models aimed at predicting booking cancellations. By preprocessing and engineering key features like booking lead time and customer demographics, the project optimizes models through various algorithms such as PCA, decision tree and random forest. The goal is to deploy a robust predictive tool that helps hotel management anticipate cancellations and optimize operational strategies accordingly.
This project explores the process of converting unstructured SMS messages into a structured format suitable for machine learning. It covers data preprocessing techniques, feature extraction, sentiment analysis and model training methodologies to build an effective spam detection system.
Various models including Decision Tree and Random Forest models are trained with a set of SMS messages, and when deployed, it predicts whether new SMS messages are spam or not. Models are designed and developed using RapidMiner.
This project delves into the intricacies of Prompt Engineering, a critical aspect of fine-tuning LLMs. It covers fundamental concepts, strategies for crafting effective prompts, and advanced techniques to optimize model performance. The experiment includes practical examples, case studies, and best practices to help users leverage prompt engineering for enhanced AI outputs in various applications.
