Loading

Pandora challenge hero

Join Pandora 2026

Challenge Summary

Build an AI-powered Retrieval-Augmented Generation (RAG) application that retrieves trusted knowledge from Pandora's ecosystem and generates reliable, evidence-based answers.

Challenge Duration: 5 HoursTheme: Inspired by Avatar: The Way of Water

Challenge Summary

Build an AI-powered Retrieval-Augmented Generation (RAG) application that retrieves trusted knowledge from Pandora's ecosystem and generates reliable, evidence-based answers.

Mission focus

Pandora Knowledge Guardian

A system that grounds every response in retrieved evidence and respects uncertainty.

Input

Documents, questions, and evidence

Output

Grounded answer with sources

Bioluminescent interface direction

Soft cyan gradients, glass panels, floating particles, and oceanic motion guide the user through the challenge.

01 / Challenge Overview

Challenge Overview

Welcome to the Pandora RAG Hackathon 2026. Your mission is to build an AI-powered Retrieval-Augmented Generation (RAG) application called the Pandora Knowledge Guardian.

Help environmental guardians, researchers, community leaders, and citizens retrieve trusted information from Pandora's knowledge base.

Generate grounded answers supported by evidence instead of relying on model memory or unsupported assumptions.

Always retrieve relevant documents before generating a response and clearly display the sources used.

02 / Problem Statement

Problem Statement

Pandora contains environmental reports, marine-life records, emergency procedures, conservation policies, community knowledge, research findings, and water quality reports distributed across many documents.

During emergencies, users struggle to quickly locate trustworthy information. Traditional AI systems may generate inaccurate or unsupported answers. Build a Retrieval-Augmented Generation application that retrieves relevant information from the knowledge base before answering user questions. Every answer should be supported by retrieved evidence and clearly indicate when sufficient information is unavailable.

03 / Your Mission

Your Mission

Participants must build an application capable of moving from Pandora documents to trustworthy, evidence-led answers without hallucination.

Upload or access Pandora documents

Bring in environmental, community, and research documents from the knowledge base.

Process documents into embeddings

Chunk source files and convert them into searchable vector representations.

Perform semantic retrieval

Fetch the most relevant passages before any answer is generated.

Generate grounded AI responses

Use the question, retrieved context, and an LLM to answer with evidence.

Display supporting evidence

Show document names, sections, excerpts, and relevance cues clearly.

Handle uncertainty without hallucinations

Say when evidence is insufficient rather than fabricating an answer.

04 / Minimum Functional Requirements

Minimum Functional Requirements

Each requirement should feel like a polished, glassmorphism-powered product surface rather than a plain checklist.

Knowledge Base Management

Environmental Reports
Marine-Life Records
Emergency Response Guidelines
Community Knowledge Documents
Conservation Policies
Research Notes
Water Quality Reports
PDFTXTCSVDOCX

Document Processing

Extract document text
Split content into meaningful chunks
Generate vector embeddings
Store embeddings inside a vector database

Semantic Retrieval

Retrieve the most relevant chunks
Use semantic search whenever the user asks a question

AI Response Generation

User Question
Retrieved Context
Large Language Model
Rely on retrieved evidence rather than model memory

Source References

Document Name
Page or Section
Retrieved Excerpt
Relevance Score (optional)

Uncertainty Handling

Clear insufficient-evidence message
Avoid unsupported claims
Prevent hallucinations

User Interface

Document Upload Area
Question Input
Generated Answer
Retrieved Sources
Reset / Clear Option
05 / Suggested Use Cases

Suggested Use Cases

Frame the solution around urgent and practical Pandora scenarios.

Environmental Incident Investigation

Track unusual conditions and surface evidence fast.

Marine-Life Protection

Protect vulnerable species and habitats with grounded guidance.

Emergency Response

Find trustworthy procedures when time is critical.

Community Knowledge Assistant

Help citizens find practical answers from local documents.

Sustainable Resource Management

Link science, policy, and action in one workflow.

Research Assistant

Compare reports, summarize findings, and expose evidence.

06 / Technology Flexibility

Technology Flexibility

Teams may use any technologies, frameworks, APIs, models, or databases they prefer. Judges evaluate the solution quality, grounded retrieval, and final experience rather than the specific stack.

Use any technology stack

Build the Pandora Knowledge Guardian with the tools that best fit your team. The challenge is about retrieval quality, grounded answers, and a polished user experience - not about forcing one framework, model, or database.

Technology agnostic
07 / Bonus Features

Bonus Features

Advanced ideas should strengthen retrieval quality or user trust.

Hybrid Search
Metadata Filtering
Query Rewriting
Result Reranking
Conversation Memory
Multiple Document Comparison
Role-Based Responses
Voice Questions
Multilingual Support
Image or Map Retrieval
Emergency Priority Classification
Confidence Scoring
Knowledge Graph Integration
Agentic RAG Workflow
08 / Final Submission

Final Submission

Use this checklist to confirm the team has everything needed for the final demo.

Working RAG Application
GitHub Repository
README
Architecture Diagram
Demo using at least three questions
Evidence of retrieved sources
3–5 Minute Final Presentation
09 / Sample Questions

Sample Questions

These questions should feel interactive, with the UI making it obvious which query is being tested.

Selected prompt

What are the possible causes of unusual changes in Pandora's ocean water?

Expected behavior

Retrieve relevant documents first.

Ground the response in evidence.

Show the source references used by the model.

Decline unsupported questions instead of fabricating an answer.

10 / Evaluation Criteria

Evaluation Criteria

A modern table keeps the judging rubric easy to scan on desktop and mobile.

Criterion
RAG Architecture & Retrieval Quality
Accuracy & Grounded Responses
Real-World Problem Solving
User Interface & Experience
Innovation
Presentation & Documentation
11 / Rules & Constraints

Rules & Constraints

The highlighted container keeps the hard requirements visually distinct from the rest of the brief.

The solution must be completed within five hours.
The application must use Retrieval-Augmented Generation.
A chatbot without document retrieval does not qualify.
Open-source frameworks, APIs, and pretrained models may be used.
Teams must disclose prebuilt components.
The final demonstration must show retrieved sources.
The application must never fabricate unsupported information.

12 / Final Mission

Final Mission

Build an intelligent knowledge system that protects Pandora's people, oceans, wildlife, and natural resources by transforming scattered information into reliable, actionable knowledge. The strongest solutions will retrieve evidence, explain reasoning, acknowledge uncertainty, and help Pandora make informed decisions.

Upload Your Git Repository

Submit your team's final GitHub repository before the challenge ends. Ensure your project is complete, accessible, and contains all required source code and documentation.

Submission Guidelines

  • Submit only one repository per team.
  • Ensure the repository is accessible to the judging panel.
  • Include all source code, required assets, and a complete README.
  • Verify that the repository link is correct before submitting.