Hybrid AI Pipelines for Compression and Retrieval in Long-Term Financial Decision-Making

This proposal aims to address a growing challenge in Canada’s financial technology sector that is to analyze ultra-long-term, unstructured financial documents in private markets for explainable decision making. Canadian investors play a significant role in the global private capital market which is estimated by industries to exceed USD $20 trillion by 2030. Unlink public markets, where data is standardized and accessible, private investments rely heavily on thousands of long and inconsistently structured documents with redundant sections and content. Analysts spend 15–20 hours per week manually extract, verify and interpret KPIs and risk metrics from these documents. Top private-equity firms begin to adopt and deploy AI for speed, and cost-effectiveness but acknowledge the risks caused by current AI models that truncate long-term data, use simple aggregation, missing critical anomalies along long trends over decades. This opacity reduces transparency and trust, hence hampers investment decisions and risk economic inefficiencies in Canada’s private market which supports Canada’s pension funds. This project advances new technical knowledge at the interaction of AI, finance and software engineering. 1) Redundancy-Aware Token Compression. A method for compressing long and redundant financial documents while preserving multi-quarter, multi -year trends and identifying anomaly essential in decision making. 2) Multi-Resolution Document Analysis Pipelines. A processing pipeline that extracts key insights using key-value query framework developed byAltQ.ai at multiple timescales including sentence-level, section-level, document-level to support trend detection and anomaly alerts over long time span of decades. 3) Financial-Domain Metrics and Benchmarking. The definition of financial-specific metrics and indexing surpass SOTA’s limitation in uniformed chucking of long documents. setting the benchmark validates over thousands of datasets. 4) Explainable and Human-AI Collaboration Design. The design that embraces explanation and human-AI collaboration ensures compliance with Canadian and global regulatory frameworks for data and AI governance in finance.

Faculty Supervisor:

Yan Liu

Student:

Partner:

AltQ.ai

Discipline:

Computer science

Sector:

Finance and Insurance

University:

Concordia University

Program: