Building enterprise AI systems across calibrated decisioning, workflow orchestration, governed authoring, and forecasting.

Focused on governed intelligence systems that combine calibrated signals, autonomous reasoning, deterministic validation, and human escalation so AI can operate safely in business-critical environments.

• AI orchestration patterns for multi-service and multi-agent workflows
• Memory and orchestration layers that consume model confidence and operational context
• Governed intelligence systems with validation, review, and traceability
• Uncertainty-aware decisioning for routing between models, LLMs, and humans
• Scalable AI architectures designed for operational deployment, not just experimentation
• Explainable enterprise AI that links technical design to measurable business outcomes

The repositories below are not disconnected experiments. Together they show a consistent architectural identity: I design enterprise AI systems that make decisions under uncertainty, expose confidence explicitly, orchestrate downstream actions, and add control layers where reliability matters.

Focus: multimodal inspection and anomaly workflows that produce calibrated confidence signals, detect fraud or outliers, and route work based on confidence instead of forcing a single model path.

Why it matters: reviewers can quickly see my approach to uncertainty-aware AI, fraud and anomaly detection, human-in-the-loop escalation, and governed multimodal systems.

Business outcome: improve decision quality while reducing the cost of blanket manual review and making escalation logic explicit.

• Calibrated Eye (Public): github.com/mkuma93/calibrated-eye
• Calibrated Eye Core (Private): github.com/mkuma93/calibrated-eye-core

Focus: orchestration layers that consume calibrated signals, maintain workflow state, and coordinate decision paths across models, tools, and human review.

Why it matters: demonstrates that I do not treat model outputs as endpoints; I design the memory and control plane that turns uncertain signals into reliable downstream action.

Business outcome: improve operational consistency, reduce brittle handoffs, and support traceable enterprise automation.

• MeMoa (Private): github.com/mkuma93/MeMoa

Focus: AI-assisted authoring workflows for compliance-heavy pharmaceutical documentation with multi-stage validation, approval, auditability, and review orchestration built in.

Why it matters: demonstrates how I structure governed document intelligence systems where correctness, traceability, and operational controls are first-class requirements.

Business outcome: reduce draft-to-release cycle time and strengthen compliance audit readiness.

• Pharma Regulatory Authoring (Public): github.com/mkuma93/pharma-regulatory-authoring
• Pharma Regulatory Authoring Core (Private): github.com/mkuma93/pharma-regulatory-authoring-core

Focus: interpretable multi-horizon forecasting for intermittent demand, with explicit occurrence and magnitude modeling, structural components, and causal sequence correction.

Why it matters: shows how I connect forecasting architecture, uncertainty, reproducible evaluation, and downstream planning rather than treating forecast accuracy as an isolated endpoint.

Business outcome: improve replenishment and planning decisions while preserving traceability from demand signals to operational action.

• SIFT — Structured Impulse Forecasting Transformer (Public): github.com/mkuma93/sift-forecasting