RAG+: A Scalable Solution for Enhancing LLM Intelligence with Comprehensive Historical Document Utilization
RAG+: A Scalable Solution for Enhancing LLM Intelligence with Comprehensive Historical Document Utilization
Participants will explore novel techniques in topic modeling, hierarchical indexing, and metadata-enhanced search to push the boundaries of LLM intelligence. The outcomes will support ADB’s digital transformation and knowledge management initiatives.
The Challenge
The Asian Development Bank (ADB) invites innovators, researchers, and AI practitioners to participate in a cutting-edge research challenge aimed at transforming the capabilities of large language models (LLMs). This challenge focuses on developing RAG+, a next-generation Retrieval-Augmented Generation system that enables global document comprehension across large-scale corpora.
Participants will explore novel techniques in topic modeling, hierarchical indexing, and metadata-enhanced search to push the boundaries of LLM intelligence. The outcomes will support ADB’s digital transformation and knowledge management initiatives.
This opportunity is under the ADB Digital Sandbox Program. Click here to learn more about available opportunities in the program.
Objectives
Participants are expected to address the following key research objectives:
-
Global Knowledge Extraction via Cross-Document Topic Modeling
Develop methods to identify latent themes and relationships across entire document collections. -
Hierarchical Document Representation and Indexing
Create scalable summarization techniques using hierarchical structures to provide global insights. -
Topic Metadata-Augmented Search
Enhance search capabilities by integrating dynamic topic metadata for improved query relevance. -
ADB-Specific Adapter Evaluation
Train and evaluate an adapter using an ADB-annotated dataset to demonstrate real-world applicability.
Expected Deliverables
Prototype of RAG+ System
A functional system showcasing advanced document comprehension and retrieval capabilities.
ADB Specific Adapter
A trained and evaluated adapter tailored to ADB’s annotated dataset.
Evaluation Report & Research Paper
A comprehensive report and a paper submitted to a flagship AI conference detailing methodology, innovations, and findings.