AI Sparkathon 2024
Problem Statements
Web Scraping and Generative AI with Vector Databases
Objectives
Participants should build an advanced AI-powered system capable of scraping complex and diverse data from the web, storing it in a vector database, and intelligently answering sophisticated user queries with high accuracy. The system should handle various data types, including text, tables, and images, and adapt to dynamically changing web environments.
Problem Statement
Develop a comprehensive and scalable solution that performs the following advanced tasks:
Advanced Web Scraping:
Scrape a wide variety of data types (text, tables, images, videos, etc.) from multiple dynamically changing websites, including those with complex structures and varied content formats.
Implement solutions to handle sophisticated anti-scraping mechanisms, including CAPTCHAs, IP blocking, and dynamic content loading (e.g., JavaScript-rendered content).
Ensure the scraping process is resilient and adaptable, capable of dynamically adjusting to different website structures and content changes without manual intervention.
Incorporate mechanisms for continuous learning to improve scraping accuracy and efficiency over time based on feedback and error detection.
Efficient Data Storage:
Store the scraped data in a vector database along with detailed meta information (e.g., source URL, document details, timestamp, content type) for efficient and contextually rich retrieval and querying.
Design and implement an advanced data organization and indexing strategy to efficiently manage, retrieve, and query diverse data types, ensuring scalability and performance even with large datasets.
Ensure data integrity, consistency, and security during the storage and retrieval processes, incorporating mechanisms for data validation and error correction.
  Intelligent Query Processing:
- Build a sophisticated RAG (Retrieval-Augmented Generation) pipeline capable of understanding, processing, and responding to nuanced and complex user queries.
- Develop advanced natural language understanding (NLU) models to accurately interpret user intents and extract relevant information from the stored data, including handling relationships and dependencies in table data.
- Integrate generative AI techniques to generate accurate, relevant, and contextually appropriate answers based on the stored data, leveraging state-of-the-art language models and fine-tuning for specific domains or query types.
Dynamic Answer Generation:
Develop a versatile answer generation mechanism capable of providing responses in various formats (text, charts, tables, images, videos, etc.) depending on the complexity and nature of the query.
Ensure the system can handle multifaceted and layered queries, synthesizing information from multiple sources and data types to provide comprehensive and insightful answers.
Implement a feedback loop to continuously improve the accuracy and relevance of the generated answers based on user interactions and feedback.
Scalability and Performance:
Design the system architecture to be highly scalable, capable of handling large volumes of data and queries with low latency and high throughput.
Incorporate performance optimization techniques to ensure efficient data processing, storage, and retrieval, minimizing resource consumption and maximizing response speed.
Implement robust monitoring, logging, and error-handling mechanisms to ensure the system’s reliability, maintainability, and fault tolerance.
Ethical and Legal Considerations:
Ensure the solution adheres to ethical standards and legal regulations related to data privacy, security, and web scraping practices.
Incorporate mechanisms for data anonymization, consent management, and compliance with relevant data protection laws (e.g., GDPR, CCPA).
AI-Powered Chart Captioning System
Objectives
Participants should build an advanced AI-powered system capable of generating detailed and complex captions for chart images uploaded by users or provided via URLs. The system will download, analyze, and describe the charts using natural language processing (NLP) techniques. Additionally, the system will include an interactive chatbot for further user interaction with the generated captions or chart images.
Problem Statement
Develop a comprehensive solution that performs the following tasks:
Chart Image Upload and URL Input
1. Provide an interface for users to upload chart images directly from their device.
- Support common image formats such as PNG and JPEG.
- Allow users to input a URL linking to a chart image, validating that the URL points to a valid image file.
- Handle both image uploads and URL inputs seamlessly, providing clear instructions and feedback to the user.
Image Downloading and Processing
- Image Downloading:
- Implement mechanisms to download images from provided URLs using reliable libraries and methods.
- Handle potential errors such as invalid URLs or unreachable links.
- Provide user feedback in case of download failures, suggesting possible reasons and corrective actions.
2. Image Processing:
- Process uploaded or downloaded images to ensure they are of suitable quality for analysis.
- Apply image processing techniques such as resizing, enhancing contrast, and correcting distortions.
- Gracefully handle low-quality or distorted images, either by improving quality or informing the user if the image is unsuitable for analysis.
Chart Identification
1. Use machine learning models or algorithms to identify the type of chart (e.g., bar, line, scatter, pie).
- Train and validate models on a diverse dataset to ensure accuracy.
- Differentiate between similar chart types and accurately identify components such as axes, labels, and legends.
Data Extraction
- Extract data from the chart image, including values represented on the axes, data points, and any labels or legends.
â—‹ Accurately interpret different scales, units, and formats of data representation in the chart.
Detailed and Complex Caption Generation
- Generate detailed and complex captions for the chart using NLP techniques. Captions should include:
â—‹ Contextual background of the dataset.
â—‹ Description of the chart type and its components.
â—‹ Analysis of key trends, patterns, and anomalies.
â—‹ Implications and conclusions drawn from the data.
â—‹ Technical details of the analysis, including statistical methods and data sources.
â—‹ Practical applications and recommendations based on the data. - Ensure the generated captions are accurate, insightful, and relevant to the audience.
Interactive Chatbot for User Engagement
- Include an interactive chatbot that users can engage with to ask questions about the generated captions or the chart image itself.
â—‹ The chatbot should provide additional explanations, clarify details in the captions, and offer further insights based on user queries.
â—‹ Leverage NLP techniques to understand and respond to user questions accurately and contextually, enhancing the user experience and providing deeper interaction with the chart data.
Virtual Try-On System
Objectives
To create an interactive platform that allows users to upload images of themselves and virtually try on clothes. The platform aims to provide a realistic 3D visualization of how the clothes fit and look on the user, enhancing the online shopping experience.
Problem Statement
Clothing Data Management:
1. Create a straightforward process to gather images and details of clothes from various sources (e.g., fashion websites, and designer catalogs).
2. Clean and standardize the clothing data.
3. Ensure accurate and high-quality 3D models of clothes.
User and Clothing Image Upload:
1. Allow users to upload their images and images of clothes they want to try on.
2. Ensure the system can handle different image formats and sizes.
3. Verify the quality and clarity of the uploaded images to ensure accurate matching.
Virtual Try-On Technology:
1. Utilize basic image processing and 3D rendering techniques.
2. Develop a model to accurately map clothes onto the user’s image.
3. Ensure the virtual try-on process realistically represents the fit, size, and appearance of
clothes on the user.
4. Generate the output where the uploaded image of the dress is matched and overlaid
onto the uploaded image of the user.
3D Visualization:
1. Implement a 3D environment where users can view themselves from multiple angles.
2. Provide features for users to adjust the view and fit of clothes.
3. Ensure the 3D visualization is responsive and user-friendly.
Dynamic Recommendation Customization:
1. Provide personalized clothing recommendations based on user interactions with the system.
2. Offer recommendations in various formats (e.g., suggested outfits, seasonal trends).
3. Handle diverse user preferences and offer a wide range of clothing options from multiple
sources.
Dynamic Dashboard Chatbot
Objectives
To develop an intelligent chatbot capable of connecting to various relational database management systems (RDBMS) with different schemas. The chatbot will dynamically generate dashboards based on user queries, providing a seamless and interactive data analysis experience.
Problem Statement
Database Connectivity:
1. Develop methods to establish secure connections to multiple types of RDBMS (e.g., MySQL, PostgreSQL, Oracle).
2. Ensure the chatbot can handle varied database schemas and structures.
3. Implement authentication and authorization mechanisms for secure access to
databases.
User Query Processing:
1. Enable the chatbot to understand and process natural language queries related to data retrieval and analysis.
2. Translate user queries into appropriate SQL commands compatible with the connected databases.
3. Handle complex queries involving multiple tables, joins, and data aggregations.
Dynamic Dashboard Generation:
1. Utilize data visualization libraries to create interactive and responsive dashboards.
2. Develop a model to interpret the context of the user’s query and determine the most
relevant data visualizations (e.g., charts, graphs, tables).
3. Ensure the generated dashboards are accurate, informative, and easy to understand.
Schema Adaptability:
1. Implement mechanisms to dynamically adapt to different database schemas.
2. Automatically detect and map the structure of connected databases to facilitate query
generation and data retrieval.
3. Handle schema changes and updates without disrupting the chatbot’s functionality.
User Interaction and Customization:
1. Provide users with options to customize their dashboards (e.g., select specific metrics, filter data).
2. Offer various formats for data visualization based on user preferences and query context.
3. Ensure a user-friendly interface for interacting with the chatbot and viewing the generated dashboards.