The Lab
Research, prototypes, and product experiments from Coreledger.

From research to product
Commercial products can be shaped by the systems, research, and experiments explored across Coreledger.
AI voice and media
Experiments where agents and media systems meet.
Fintech AI
Research and prototypes from our background in post-trade systems and reconciliation.
Bond reconciliation / post-trade break detection
ResearchAutomated exception handling and reconciliation for fixed-income post-trade workflows.
View on GitHubVoice-to-trade / confirmation research
ResearchOpen-source research prototype for turning spoken bond-trade conversations into structured confirmations and audit records.
View on GitHubWhite paper
ResearchMusodza, K. (2025). Bond Settlement Automated Exception Handling and Reconciliation. Zenodo.
Read on ZenodoOpen models and experiments
Research models we have published openly on Hugging Face.
DistilBERT-Reconciler
LiveFine-tuned DistilBERT on 3.2M labeled post-trade break descriptions and resolution actions.
View on Hugging FaceFails Forecaster
LivePredicts next-day settlement-fail notional for US Treasuries and corporates using XGBoost + LightGBM.
View on Hugging FaceSettlement Stress Flagger
LiveFlags days where settlement fails are in the top 10% of historic values for proactive remediation.
View on Hugging FaceBERT-Breaks baseline
Coming soonVanilla BERT baseline for exception handling and reconciliation. Training in progress.
View on Hugging FaceDeveloper tools
Tools and adapters that support governed agent workflows.
Contextus templates
LiveReusable policies and workflow patterns for governed agent actions. Drop-in starting points for teams.
Explore ContextusMCP experiments
BetaAdapter and host integrations for agent tools that need governance, approval, and audit.
Agent workflow tools
ResearchHelpers for IDE, CI, and editor-native agent workflows that route through Contextus governance.
Explore the work behind the ideas.
Much of our research, models, adapters, and experiments are available through GitHub and Hugging Face.