AI Portfolio
Eight Fronts of Applied AI
The AI Transformation page explains the philosophy; this page is the inventory. Each domain is tagged by its real status — in daily use, in design, or under exploration — because an honest portfolio beats an inflated one.
AI Systems
The assembled working system: model, prompt library, verification step, and filing routine — treated as one controlled unit, not scattered chats.
- Daily reporting system
- Investigation write-up system
- Reconciliation support
AI Agents
Evaluating agent-style tooling for multi-step tasks — collection-note triage, follow-up drafting — with hard rules on what an agent may never decide alone.
- Task decomposition patterns
- Human approval gates
- Tool-use safety rules
Automation
Converting repetitive checking and reporting into assisted flows: model drafts, human verifies, output files identically every time.
- Report drafting
- Log structuring
- Checklist generation
Prompt Engineering
Prompts written like controlled procedures: fixed inputs, explicit steps, expected output format, and refusal instructions for missing data.
- Reusable template library
- Structured output formats
- Bilingual AR/EN prompting
AI Architecture
Designing where a model, a rule, and a human each belong in a pipeline — capture, structure, validate, report — before any tool is chosen.
- Financial middleware design
- Exception-queue pattern
- Audit-trail-first thinking
AI Workflows
End-to-end working routines in production use: field notes to management report, observation log to incident narrative.
- Daily report workflow
- Incident narrative workflow
- Summary-of-summaries reviews
Local LLM
Testing on-device models for one reason: financial data that should never leave the building. Privacy is the requirement, not the trend.
- On-device inference
- Anonymization-first policy
- Sensitive-data workflows
Business AI
The translation layer: turning AI capability into operations language — what it saves, what it risks, and how colleagues adopt it without fear.
- Colleague enablement
- Adoption by usefulness
- AI-with-accountability principles