Projects
Applied AI systems for finance and risk.
These projects connect accounting policy, credit risk, geospatial methods and the practical realities of regulated financial institutions.
Featured Project
Ile Owo
An eight-agent IFRS advisory system.
Ile Owo, meaning "house of money" in Yoruba, is a multi-agent accounting policy advisor built on the Anthropic Claude API. Rather than a single chatbot, it mirrors how a real policy team works.
The architecture includes an Orchestrator, Filter, Summaryan, Historian, Insider, Outsider, Auditor and Scribe. Sources span the UK Endorsement Board, BIS, Big 4 technical libraries and internal repositories.
The architecture expanded from an earlier five-agent prototype to capture the adversarial review structure that gives professional policy advice its rigour. Built in Python on Ubuntu. Active development.
Architecture
The eight agents of Ile Owo.
The Orchestrator routes each enquiry through three stages — sourcing, adversarial review and synthesis — to produce a defensible policy answer.
Agent roles
- Orchestrator — routes work between stages.
- Filter — scopes the question to relevant sources.
- Summaryan — extracts the load-bearing claims.
- Historian — situates the issue in standard-setting history.
- Insider — argues the preparer's position.
- Outsider — argues the auditor/regulator's position.
- Auditor — resolves the exchange against the standard.
- Scribe — produces the final advisory note.
Hackathon Project
Geospatial Risk Analysis
Collateral value and balance sheet resilience.
This project applied geospatial analysis to credit risk. It used location-aware data to assess collateral value and model balance sheet resilience, combining geographic data science methods with traditional credit risk frameworks.
The work explored how spatially correlated shocks, including climate, regional economic stress and infrastructure risk, propagate through a bank's collateral portfolio in ways that aggregate-level models tend to miss.