GENERATIVE AI PROJECTS
.png)
Inventory Insights Generator using Google PaLM
Crafted a sophisticated query system that empowers users to generate insights from thier inventory using plain English. This innovative solution seamlessly integrates Google PaLM for SQL query generation, few-shot training, HuggingFace for text embeddings, and Chroma as the vector database. With this orchestration, the application autonomously executes SQL queries on relational inventory tables, delivering immediate, context-rich results for enhanced data-driven decision-making across diverse domains.
Lecture Loom: Interactive Learning Platform
Engineered an innovative application for Vanderbilt University, facilitating student learning by enabling seamless interaction with lecture videos and audio files. Leveraged OpenAI Whisper to transcribe uploaded video/audio content into text format and Pinecone vector database for efficient storage and retrieval of text embeddings. This RAG model deployed using the Serverless Framework, integrates with AWS Lambda, S3 buckets and API Gateways to provide students with relevant context and answers using GPT-4, optimizing learning experiences.
Multilingual AI-powered Chatbot for Customer Support
Developed a cutting-edge multilingual AI-powered chatbot for customer support at One Energy, North America's largest wind energy supplier. This innovative solution will significantly enhance customer engagement, streamline support processes, and reduce response times, resulting in an improvement in customer satisfaction and operational efficiency. The chatbot's seamless multilingual capabilities will further expand One Energy's reach and service accessibility, cementing its position as an industry leader.
ArticleSense: AI Driven News Query WebApp
Deployed an advanced web application that utilizes Langchain and FAISS (Facebook AI Similarity Search) to answer user questions based on news articles provided via URLs. The project involved fine-tuning the model with few-shot training, resulting in a highly accurate and versatile system. This empowers users to gain instant, contextual insights from news content, enhancing their information retrieval experience and expanding access to valuable news sources.
.png)
MorganTales: Your Life, Narrated in the style of Morgan Freeman
Implemented a fun project where users can upload any video, utilizing GPT Vision models to dynamically generate narrations reminiscent of the iconic style of Morgan Freeman. This project seamlessly translates visual content into compelling life narratives. The system empowers users to experience personalized storytelling, showcasing a fusion of cutting-edge technologies to offer a unique and engaging way for individuals to narrate their life stories through uploaded videos. Because, let's face it, who wouldn't want their life story narrated by Morgan Freeman?
DocVision: Empowering Queries on Invoices and Prescriptions
Developed a web app to extract actionable insights from uploaded invoices and prescriptions, leveraging the cutting-edge capabilities of Gemini Pro Vision. The LLM parses and interprets the content, extracting pertinent information such as billing details, medication prescriptions, and patient details. This streamlined workflow empowers users to pose questions in simple English and derive actionable insights from the uploaded documents. This enhances user experience, data accessibility and operational efficiency in healthcare and finance domains.
.png)
AdPulse: Generate short Ads for your Product
Leveraged the capabilities of the open-source Large Language Model, Bloom 1b7, through a meticulous fine-tuning process using the Parameter Efficient Fine Tuning (PEFT) methodology. Conducted comprehensive training on a diverse dataset, encompassing product descriptions and ads, resulting in the model's proficiency in generating compelling short advertisements based on input product names and descriptions. The fine-tuned model was seamlessly deployed on the Hugging Face platform