Designed and developed an AI-powered chat application which aimed at efficiently managing Calendly events through a user-friendly chat interface.

• Played with Langchain for the first time. Langchain's Function Calling feature enabled seamless integration of external tools and APIs, like Calendly, with the language model, allowing efficient management of Calendly events through a user-friendly chat interface.
• Dmplemented complex features like authentication in Python for the backend. Working with Python and integrating third-party APIs like Calendly required a solid understanding of backend development principles and best practices.
• Developed a responsive frontend using React and TailwindCSS. Building a user-friendly chat interface with optimal performance and responsiveness across devices was crucial for delivering an excellent user experience.

Source code is available on Github

I built an AI agent called partiful.ai, which is an intelligent tool aimed at helping developers to get started with the hackathon by helping them to ask relevant questions about different new AI tools and how to integrate them for their use case. It does this by scraping official documents of the AI tools like Llama Index, Fireworks, etc and creating a RAG on it. With a tight deadline, I still managed to create a fully functional prototype. By leveraging cutting-edge GPT-4 technology integrated with a powerful Llama Index and Fireworks, we engineered a full functional solution. While the performance of the agent fell short of our expectations, I am confident that with future work, we can make significant improvements. One of the primary challenges in this project was integrating scraped data with its source to enhance credibility and provide users with access to the original documents. To tackle this, I first delved into how LlamaIndex utilizes Documents/Nodes to store data chunks. Then, I modified the existing logic to incorporate storing the source link as metadata within the Document.

I build Fiction Lens, a tool for fiction readers and writers to analyze stories by leveraging a chat interface to answer questions about dialogue, plot, and visualizations like story arcs. I used LlamaIndex to create indexes for our custom data and stored them in our vector database provided by Astra DB by DataStax. Here are few of the learnings from the development journey of building Fiction Lens:

• Played with LlamaIndex for the first time. LlamaIndex leverages retrieval-augmented generation (RAG) to build specialized inverted indexes optimized for vector similarity search, enabling efficient lookups for custom data sources like paragraphs from the novel indexed in our vector database.
• Played with vector database Astra DB for the first time. It is used for storing and querying the inverted indexes created by LlamaIndex, enabling fast similarity searches to power real-time responses in our generative AI application.
• Working on a tight schedule during the hackathon taught me the importance of efficient time management. It was crucial to prioritize tasks and allocate time effectively to meet deadlines and deliverables.

Source code is available on Github

I built an innovative AI agent, creativ.ai, which aimed to streamline the process of generating personalized content, particularly targeting founders, influencers, and individuals who seek to minimize the time spent on crafting social media posts. With a tight three-hour deadline, I managed to create a fully functional prototype. Leveraging the cutting-edge technologies of GPT-4 and DALL·E 2, integrated within our React backend, I engineered a solution that exceeded our expectations.

I built an innovative AI agent called Insightly.ai, which is an intelligent tool aimed at helping developers and project managers pick the right dataset based on their requirements. It does this by allowing users to query the agent with specific questions to better understand the available data. With a tight deadline, I still managed to create a fully functional prototype. By leveraging cutting-edge GPT-4 technology integrated with a powerful Hex tool, I engineered a full functional solution. The key learning experience for me from this hackathon was discovering how to quickly ramp up on new tools like Hex to build projects under pressure. Hex integrates the best features of notebooks, and data platforms like Snowflake, MySql etc into one environment. This speeds up and improves data related work for teams - from exploration to productionisation.

Feel free to try out the Demo

A comprehensive MERN (MongoDB, Express.js, React, Node.js) stack application designed to facilitate support ticket management. It incorporates user authentication, ticket creation, updates, and notes functionality. The main functionalities of the project are :-

• User Authentication: Users can register, log in, and log out securely. Authentication is implemented using JWT (JSON Web Tokens) for protected routes.
• Ticket Management: Users can create, view, update, and close support tickets. The app enables users to add notes to existing tickets for better communication and issue resolution.

Feel free to try out the Live Support Desk App and explore its source code on Github

The House Marketplace App is a Progressive Web App (PWA) designed and developed using React and Tailwind CSS. It serves as a platform for streamlined property listing, whether for rental or sale of residential properties. This documentation outlines the primary features and architecture of the app. The main functionalities of the project are :-

• User Authentication The app provides user authentication through Google Auth, ensuring secure access to the platform.
• Property Listings Users can create, edit, and manage property listings with details such as images, property type, location, price, and description.
• Geolocation: Geolocation features help users find properties near their current location, enhancing the user experience.
• Dynamic Animations: The app includes dynamic animations for a visually engaging and interactive user interface.
• Payment Integration: Stripe is integrated for payment processing, allowing users to make secure payments for property listings and services.
• Real-time Updates: Firebase backend is utilized for secure data storage and real-time updates, ensuring a responsive and dynamic user interface.

Feel free to try out the Live House Marketplace App and explore its source code on Github

It offers essential functionalities for searching and viewing information from GitHub profiles, including recent repositories. The application leverages GitHub APIs for accessing user data and incorporates advanced features like state and effect hooks, context, and a reducer for state management. The provides the following primary functionalities :-

• User Search: Users can search for a specific GitHub user by entering their username. The app utilizes the GitHub API to fetch user data.
• User Profile Display: The app displays essential information from the user's GitHub profile, including their name, bio, location, avatar, and a link to their GitHub profile.
• Recent Repositories: Users can view a list of recent repositories from the user's profile, along with important details such as repository name, description, and the number of stars and forks.

Feel free to try out the Live Github Finder App and explore its source code on Github

Worked on working my first React project to collect and manage feedback. The main functionalities of the project are :-

• Add New Feedback: Users can add new feedback, which includes a rating and a comment.
• Delete Existing Feedback: Users can delete feedback entries.
• List All Existing Feedbacks: The app displays a list of all existing feedback entries, allowing users to browse through them.
• Display Average Rating: The application calculates and displays the average rating of all feedback entries.
• Edit Existing Feedbacks: Users can edit the existing feedback entries, including both the rating and comment.

Feel free to try out the Live Feedback App and explore its source code on Github

Built an application to improve communication medium for visually impaired people by recognizing their hand gestures.

• Implemented Convex Hull Algorithm to extract hand gestures from video and to classify it in real time.
• Experimented with various Neural Network algorithms to improve the accuracy and performance of Hand Gesture Recognition system.
• Created Convolutional Neural Network based model for classifying hand gestures and achieved 97% accuracy.
• Cited the findings in the paper Transition from Convex Hull Algorithm to CNN for Image Classification at IRJET

Source code is available on Github