HackSpark 2025

Generative AI

1. LLM Document Processing System
   Build a system that uses Large Language Models (LLMs) to process natural language queries and retrieve relevant information from large unstructured documents such as policy documents, contracts, and emails.
   
   Objective
   
   The system should take an input query like:
   "46-year-old male, knee surgery in Pune, 3-month-old insurance policy"
   
   It must then:
   
   

1. Parse and structure the query to identify key details such as age, procedure, location, and policy duration.

2. Search and retrieve relevant clauses or rules from the provided documents using semantic understanding rather than simple keyword matching.

3. Evaluate the retrieved information to determine the correct decision, such as approval status or payout amount, based on the logic defined in the clauses.

4. Return a structured JSON response containing: Decision (e.g., approved or rejected), Amount (if applicable), and Justification, including mapping of each decision to the specific clause(s) it was based on.

2. Requirements
   
   

1. Input documents may include PDFs, Word files, or emails.

2. The query processing and retrieval must work even if the query is vague, incomplete, or written in plain English.

3. The system must be able to explain its decision by referencing the exact clauses used from the source documents.

4. The output should be consistent, interpretable, and usable for downstream applications such as claim processing or audit tracking.

Applications

This system can be applied in domains such as insurance, legal compliance, human resources, and contract management.

Sample Query

"46M, knee surgery, Pune, 3-month policy"

Sample Response

{

  "decision": "approved",

  "amount": 50000,

  "justification": "Knee surgery is covered under Clause 3.2 of the policy document for policies active for at least 90 days."

}

3.  Focus Areas: Natural Language Processing, Document Analysis, Semantic Search.
   
   

4. AI-Powered Full-Stack Website & Backend Builder for Non-Technical Founders
   
   Background
   
   In the startup world, countless early-stage founders and entrepreneurs have brilliant ideas but struggle to build and launch their products due to a lack of technical knowledge. Current no-code tools like Framer or Webflow allow drag-and-drop designs but do not provide full ownership of frontend and backend code. On the other hand, AI code generators like v0 and Bold lack a visual canvas or do not support end-to-end functionality for real deployment. There is a growing need for a platform that lets users design visually, generate real code, build backend logic, and deploy the full project, all with the power of AI.
   
   The Challenge
   
   Build a prototype of a platform that enables non-technical users to create and deploy full-stack web applications using drag-and-drop tools combined with AI-assisted code generation.
   
   The platform should allow users to:
   
   

1. Design the frontend visually using a canvas or block system.

2. Automatically generate production-ready code (HTML/CSS/JS or React).

3. Generate backend APIs, database models, and authentication flows using AI.

4. Preview, edit, and deploy the full application — frontend, backend, and database — with minimal effort.

5. Optionally integrate features like AI chat assistant, role management, test generation, and component upload.

5. Your Objective
   
   As a team, you are tasked with building an MVP (Minimum Viable Product) of this platform. The solution should focus on simplicity, usability, and real functionality — especially for people with zero coding knowledge.
   
   Bonus Features to Explore (Optional but Impressive)
   
   

1. AI chat assistant to help with edits.

2. Database schema visualizer and backend logic builder.

3. One-click deployment using platforms like Vercel, Render, or Supabase.

4. Ability to upload code snippets and convert them into visual components.
   Focus Areas: No-Code Platforms, AI Code Generation, Full-Stack Development.

6. Space Station Hackathon: Object Detection with Synthetic Data
   Duality AI presents a challenge to train an AI model for object detection in a space station environment using synthetic data from Duality AI’s Falcon digital twin simulation platform. Participants will develop skills in AI training, enhance their portfolios, and connect with Duality AI and peers while competing for prizes and recognition.
   
   Objectives
   
   

1. Train a robust object detection model using the provided synthetic dataset to accurately detect and classify space station objects (Toolbox, Oxygen Tank, Fire Extinguisher) — critical for operational safety.

2. Evaluate model performance across challenging scenarios, including varied lighting, object angles, and occlusions.

3. Benchmark and optimize the model for accuracy, generalizability, and efficiency in real-world deployment scenarios.

4. (Bonus) Create an application that uses the detection model and describe a plan to keep the model up-to-date using Falcon.

7. Importance of Digital Twins in Space Station Simulation
   
   Digital twins are highly realistic virtual representations of real-world systems. This challenge uses data from a digital twin space station environment generated within Falcon. Digital twins enable:
   
   

1. Simulation of real-world conditions (e.g., lighting, occlusions).

2. Training AI models without real-world data collection.

3. Reproducible and scalable experimentation.

8. Hackathon Structure
   
   Data Overview
   
   

1. Participants will work with a dataset generated from FalconEditor, featuring a simulated space station scene with:

• Varying lighting conditions.

• Occlusions and object overlap.

• Diverse camera angles and distances.

2. Target Object Categories: Toolbox, Oxygen Tank, Fire Extinguisher.

9. Participants will process synthetic data, train an AI model to classify these objects, and optimize detection accuracy under realistic constraints.
   
   Hackathon Tasks
   
   To maximize efficiency, teams should divide responsibilities:
   
   

1. AI Engineering

• Train and fine-tune the YOLOv8 object detection model using the generated dataset.

• Evaluate model performance.

• Use optimization techniques to improve accuracy and reduce inference time.

2. Documentation & Presenting Final Team Results

• Document the workflow, including dataset generation, model training, and evaluation metrics.

• Prepare a structured report and final presentation showcasing findings, results, and insights.

• Create visualizations such as confusion matrices and performance graphs to highlight model behavior.

3. (Bonus) Create an App

• Create a computer or phone app that uses the model.

• Describe the plan for keeping the model up-to-date using Falcon.

10. Key Deliverables
    
    At the end of the hackathon, teams must submit:
    
    

1. Trained Object Detection Model

• Fully trained YOLOv8 model that detects and classifies all three object categories.

• Include model weights, scripts, and config files.

2. Performance Evaluation & Analysis Report

• mAP Scores (@0.5 IoU) to evaluate model accuracy.

• Confusion Matrix to visualize class-wise performance.

Failure Case Analysis.
Focus Areas: Object Detection, Synthetic Data, Digital Twins, Model Optimization.

Cybersecurity

7. AI-Driven Phishing Detection for Small Businesses
   
   Background
   
   Small businesses in India are increasingly targeted by phishing attacks via emails, SMS, and social media, leading to data breaches and financial losses. Existing anti-phishing tools are often too complex or expensive for small-scale enterprises.
   
   The Challenge
   
   Develop an AI-powered platform to detect and prevent phishing attacks for small businesses. The platform should:
   
   

• Use generative AI to analyze incoming emails, SMS, or social media messages for phishing indicators (e.g., suspicious links, language patterns).

• Provide real-time alerts to users with clear explanations of detected threats.

• Offer a dashboard to track and manage potential threats.

• Educate users on phishing prevention through an AI-driven chatbot.

8. Your Objective
   
   Build an MVP that protects small businesses from phishing attacks while being affordable and user-friendly.
   
   Bonus Features
   
   

• Multilingual support for Indian languages to cater to diverse users.

• Integration with popular email and messaging platforms (e.g., Gmail, WhatsApp).

• Automated response suggestions to report phishing attempts.
  Focus Areas: Phishing Detection, Natural Language Processing, Real-Time Analytics, User Education.

9. Secure Remote Work Authentication System
   
   Background
   
   With the rise of remote work in India, ensuring secure access to company systems is critical. Weak authentication methods lead to unauthorized access, data leaks, and operational disruptions, especially for SMEs with limited cybersecurity budgets.
   
   The Challenge
   
   Create an AI-powered authentication system for secure remote access to workplace systems. The system should:
   
   

• Use generative AI to verify user identities through multi-factor authentication (e.g., biometrics, behavioral analysis).

• Detect anomalies in login patterns (e.g., unusual locations or times).

• Provide a user-friendly interface for employees to authenticate securely.

• Ensure compliance with Indian cybersecurity regulations.

10. Your Objective
    
    Develop an MVP that enhances security for remote workers while maintaining ease of use.
    
    Bonus Features
    
    

• Integration with existing workplace tools (e.g., Slack, Microsoft Teams).

• Offline authentication options for low-connectivity environments.

• AI-driven risk scoring for login attempts.
  Focus Areas: Authentication, Behavioral Analysis, Cybersecurity, Regulatory Compliance.

Web3

9. Decentralized Supply Chain Transparency Platform
   
   Background
   
   Supply chain fraud and lack of transparency are major issues in India, particularly in agriculture and manufacturing, leading to inefficiencies and consumer distrust.
   
   The Challenge
   
   Build a Web3-based platform using blockchain to enhance transparency in supply chains. The platform should:
   
   

• Use smart contracts to record and verify supply chain data (e.g., origin, processing, transport).

• Allow consumers to trace product journeys via a user-friendly interface (e.g., QR code scanning).

• Integrate generative AI to summarize supply chain data and generate trust reports for consumers.

• Ensure data security and immutability using blockchain technology.

10. Your Objective
    
    Create an MVP that promotes trust and transparency in supply chains, with a focus on agriculture or small-scale manufacturing.
    
    Bonus Features
    
    

• Multilingual interface for Indian languages.

• Integration with IoT devices for real-time supply chain tracking.

• Rewards system for ethical suppliers using cryptocurrency tokens.
  Focus Areas: Blockchain, Smart Contracts, Supply Chain Transparency, Generative AI.

11. Web3-Based Freelancer Payment System
    
    Background
    
    Freelancers in India often face delayed payments, high transaction fees, and disputes over work agreements. Traditional payment systems lack the flexibility and trust needed for the growing gig economy.
    
    The Challenge
    
    Develop a Web3-based platform for secure and transparent freelancer payments. The platform should:
    
    

• Use smart contracts to automate payments upon work completion.

• Integrate generative AI to generate fair work agreements based on project descriptions.

• Provide a decentralized escrow system to ensure trust between freelancers and clients.

• Support cryptocurrency payments to reduce transaction fees.

12. Your Objective
    
    Build an MVP that streamlines payments and builds trust in the gig economy.
    
    Bonus Features
    
    

• AI-driven dispute resolution system.

• Multilingual support for Indian freelancers.

• Integration with existing freelance platforms (e.g., Upwork, Fiverr).
  Focus Areas: Blockchain, Smart Contracts, Gig Economy, Generative AI.

HealthTech

4. AI-Driven Telemedicine Accessibility Platform
   
   Background
   
   In rural and underserved areas of India, access to quality healthcare is limited due to a lack of medical professionals and infrastructure. Many individuals struggle to get timely medical advice, especially for non-emergency conditions, leading to delayed treatments and worsened health outcomes.
   
   The Challenge
   
   Build an AI-powered telemedicine platform that enables patients to describe symptoms in their native language (text or voice input) and receive accurate preliminary diagnoses and recommendations. The platform should:
   
   

• Use generative AI to interpret symptoms described in natural language, including regional dialects.

• Suggest possible medical conditions and immediate care steps, referencing credible medical guidelines.

• Connect patients with nearby healthcare providers or teleconsultation services.

• Provide a multilingual chatbot for follow-up queries and health education.

5. Your Objective
   
   Create an MVP that prioritizes accessibility, supports Indian languages, and ensures usability for low-literacy users.
   
   Bonus Features
   
   

• Integration with wearable devices to monitor vital signs.

• Offline functionality for areas with poor internet connectivity.

• Visual guides for basic first-aid procedures.
  Focus Areas: Telemedicine, Multilingual AI, Healthcare Accessibility, Natural Language Processing.

6. Smart Medication Waste Reduction System
   
   Background
   
   Expired or unused medications contribute significantly to medical waste, posing environmental risks and financial losses for households and pharmacies. In India, improper disposal of medications is a growing concern, with limited systems to track or manage medication lifecycles.
   
   The Challenge
   
   Develop an AI-powered system to reduce medication waste by tracking and managing medication inventories for households or small pharmacies. The system should:
   
   

• Use image recognition to scan and identify medications (pills, packaging, or labels).

• Track expiration dates and send reminders for timely use or disposal.

• Suggest donation options for unused, unexpired medications to nearby NGOs or healthcare facilities.

• Generate safe disposal guidelines compliant with local regulations.

7. Your Objective
   
   Build an MVP that helps users manage medications efficiently, reduces waste, and promotes sustainable practices.
   
   Bonus Features
   
   

• Integration with local pharmacy databases for real-time stock updates.

• AI chatbot to educate users on proper medication storage.

• Gamification to encourage sustainable behavior (e.g., rewards for proper disposal).
  Focus Areas: Medical Waste Management, Image Recognition, Sustainability, User Engagement.

FinTech

6. AI-Powered Financial Literacy Assistant for Rural Communities
   
   Background
   
   Financial illiteracy is a major barrier in rural India, where many individuals lack access to basic financial education, leading to poor financial decisions, debt traps, and low savings.
   
   The Challenge
   
   Create an AI-powered mobile application that delivers personalized financial education to rural users in their native languages. The system should:
   
   

• Use generative AI to create easy-to-understand financial advice based on user inputs (e.g., income, expenses, goals).

• Provide interactive tutorials on budgeting, saving, and avoiding financial scams.

• Offer voice-based interactions for low-literacy users.

• Suggest local financial products (e.g., micro-savings accounts, government schemes).

7. Your Objective
   
   Develop an MVP that empowers rural users with financial knowledge and promotes financial inclusion.
   
   Bonus Features
   
   

• Gamified learning modules to increase engagement.

• Integration with UPI for seamless transaction tracking.

• Offline mode for areas with limited connectivity.
  Focus Areas: Financial Inclusion, Multilingual AI, User Education, Voice AI.

8. Fraud Detection for Small-Scale Digital Transactions
   
   Background
   
   With the rise of digital payments in India, small-scale merchants and individuals are increasingly targeted by fraudsters. Existing fraud detection systems are often designed for large-scale banking, leaving small transactions vulnerable.
   
   The Challenge
   
   Build an AI-powered system to detect and prevent fraud in small-scale digital transactions (e.g., UPI, mobile wallets). The system should:
   
   

• Use generative AI to analyze transaction patterns and flag suspicious activities in real-time.

• Provide user-friendly alerts and explanations for flagged transactions.

• Offer merchants tools to verify customer identities securely.

• Ensure privacy and compliance with Indian data protection regulations.

9. Your Objective
   
   Create an MVP that enhances trust in digital payments for small merchants and individuals.
   
   Bonus Features
   
   

• Voice-based fraud alerts for accessibility.

• Integration with local payment platforms like UPI or Paytm.

• Educational pop-ups to teach users about fraud prevention.
  Focus Areas: Fraud Detection, Real-Time Analytics, Cybersecurity, Financial Trust.

Space Station Hackathon: Object Detection with Synthetic Data 

15. Space Station Hackathon: Object Detection with Synthetic Data
    Duality AI presents a challenge to train an AI model for object detection in a space station environment using synthetic data from Duality AI’s Falcon digital twin simulation platform. Participants will develop skills in AI training, enhance their portfolios, and connect with Duality AI and peers while competing for prizes and recognition.
    
    Objectives
    
    

    1. Train a robust object detection model using the provided synthetic dataset to accurately detect and classify space station objects (Toolbox, Oxygen Tank, Fire Extinguisher) — critical for operational safety.

    2. Evaluate model performance across challenging scenarios, including varied lighting, object angles, and occlusions.

    3. Benchmark and optimize the model for accuracy, generalizability, and efficiency in real-world deployment scenarios.

    4. (Bonus) Create an application that uses the detection model and describe a plan to keep the model up-to-date using Falcon.

16. Importance of Digital Twins in Space Station Simulation
    
    Digital twins are highly realistic virtual representations of real-world systems. This challenge uses data from a digital twin space station environment generated within Falcon. Digital twins enable:
    
    

    1. Simulation of real-world conditions (e.g., lighting, occlusions).

    2. Training AI models without real-world data collection.

    3. Reproducible and scalable experimentation.

17. Hackathon Structure
    
    Data Overview
    
    

    • Varying lighting conditions.

    • Occlusions and object overlap.

    • Diverse camera angles and distances.

    • Participants will work with a dataset generated from FalconEditor, featuring a simulated space station scene with:

    • Target Object Categories: Toolbox, Oxygen Tank, Fire Extinguisher.

18. Participants will process synthetic data, train an AI model to classify these objects, and optimize detection accuracy under realistic constraints.
    
    Hackathon Tasks
    
    To maximize efficiency, teams should divide responsibilities:
    
    

    • Train and fine-tune the YOLOv8 object detection model using the generated dataset.

    • Evaluate model performance.

    • Use optimization techniques to improve accuracy and reduce inference time.

    • Document the workflow, including dataset generation, model training, and evaluation metrics.

    • Prepare a structured report and final presentation showcasing findings, results, and insights.

    • Create visualizations such as confusion matrices and performance graphs to highlight model behavior.

    • Create a computer or phone app that uses the model.

    • Describe the plan for keeping the model up-to-date using Falcon.

    • AI Engineering

    • Documentation & Presenting Final Team Results

    • (Bonus) Create an App

19. Key Deliverables
    
    At the end of the hackathon, teams must submit:
    
    

    • Fully trained YOLOv8 model that detects and classifies all three object categories.

    • Include model weights, scripts, and config files.

    • mAP Scores (@0.5 IoU) to evaluate model accuracy.

    • Confusion Matrix to visualize class-wise performance.

    • Failure Case Analysis.
      Focus Areas: Object Detection, Synthetic Data, Digital Twins, Model Optimization.

    • Trained Object Detection Model

    • Performance Evaluation & Analysis Report

Open Innovation Track 

Background
Innovation is not bound by a single domain. From solving everyday challenges to creating futuristic solutions, great ideas often emerge when there are no limits. The Open Innovation Track gives participants complete freedom to explore any sector—healthcare, environment, education, finance, mobility, space, entertainment, or entirely new fields.

The Challenge
Build a technology-driven solution that tackles a real-world problem or creates a new opportunity in any domain of your choice. Your project should:

• Address a clearly defined need or gap in the chosen sector.

• Use innovative thinking and, where possible, emerging technologies such as AI, Blockchain, IoT, XR, or Robotics.

• Be feasible to demonstrate as a prototype or MVP within the hackathon.

• Have potential for real-world adoption and scalability.

Your Objective
Develop a working solution (prototype/MVP) that showcases originality, impact, and technical execution—whether it’s a tool, platform, service, or product. The end goal is to demonstrate an idea that could inspire change, disrupt an industry, or significantly improve lives.

Bonus Features

• Cross-domain innovation combining two or more sectors.

• Sustainability-focused or socially impactful solutions.

• Accessibility features for differently-abled or low-literacy users.

• Offline functionality for regions with limited connectivity.

• Integration with open datasets or public APIs.

Focus Areas (Examples, Not Limits)

• Healthcare & Wellness

• Environment & Sustainability

• Education & Skill Development

• FinTech & Financial Inclusion

• Smart Cities & Mobility

• Cybersecurity & Privacy

• Space Tech & Exploration

• Entertainment & Media Innovation