This competition challenges participants to build an end-to-end AI system that generates structured, high-quality presentations from prompts using retrieval-augmented generation (RAG) over a provided document corpus. Participants will design pipelines that retrieve relevant information from a curated PDF knowledge base, generate coherent slide content, and structure outputs into presentation-ready formats.
This is an individual, time-bound evaluation (48 hours per participant) designed to assess real-world AI system design skills. Top-performing participants may be considered for internships and collaboration opportunities on INICAI-driven projects.
The goal of this competition is to build a system that converts a given prompt into a structured presentation using only the provided dataset, leveraging RAG pipelines and open-source tools. Your solution should: - Retrieve relevant context from documents - Generate accurate and concise content - Organize information into clear slide structures
Modern AI systems are increasingly used to automate knowledge synthesis tasks such as report writing and presentation generation. However, generating factually grounded, well-structured presentations requires effective information retrieval, strong summarization and reasoning, and clear structural formatting. In this competition, you will work with a curated dataset consisting of research papers (e.g., arXiv), public reports, Wikipedia-derived documents, and document layout data.
Your task is to design a pipeline that integrates document retrieval, context-aware generation, and structured output formatting.
Submissions will be evaluated based on: 1. Content Relevance: Accuracy of retrieved and generated information and alignment with the given prompt. 2. Structure Quality: Logical organization of slides and clarity of titles and bullet points. 3. Coverage: Completeness of the topic and inclusion of key aspects. 4. Grounding (RAG Effectiveness): Use of relevant document sources and reduction of hallucinated content.
For each prompt in the test set, you must generate a JSON file containing generated slides with the following structure:
{
"slides": [
{
"title": "Slide Title",
"bullets": ["Point 1", "Point 2"],
"source": "document_id"
}
]
}
Preference will be given to efficient, scalable, and well-designed pipelines. Clear documentation and reproducibility are essential for top performers, as this competition emphasizes practical system design over model complexity.