Rendered systems a reader can inspect.
A simple map of AI-assisted finance risk work.
Use this when you want the shortest serious read: what Bionic Banker is, what runs, what records support it, what stays private, and what to check next.
AI agents for finance risk. Simple records first, human control always.
Bionic Banker connects wallet-risk review, fraud and AML case work, agent workflows, and public writing. The rule is simple: show what happened, show what was checked, and keep the final decision with a person.
6 inspectable systems, one map.
Score, rule explanation, and plain-language note for sample wallet-risk rows.
AML Status AuditabilitySource trail, checks, visible claims, and explicit clear limits.
Fraud Alert Triage Workflow Alert operationsSynthetic signals converted into a system record, case note, and human next step.
Agent Workflow Case Review Workflow reviewOne messy finance-risk workflow converted into a case record, review summary, and human review step.
Agent Chess / Agent Workflow Agent operationsTurn-taking, state preservation, review gates, and public records without outside-action power.
Site Health / Health Board Self-checking recordsMetric contracts, link checks, JSON-framing checks, mobile QA, and public status.
The public record map is counted and limited.
Reader-facing sections with sources and boundaries.
Reader-facing sources linked from the evidence catalog.
Explicit limits that prevent overclaiming.
Writing that supports the finance, AI, and agent-system context.
Built from simple pieces a reader can recognize.
- Astro static site with committed static output
- GitHub repositories for code, issues, and public change history
- Agent runtime for tool use, review records, and bounded workflows
- Codex, Claude Code, and OpenCode-style coding agents tested as builders and reviewers
- local models for low-cost reasoning tests
- SQLite ledgers and queues for small-system records
- Node.js QA checks for copy, links, metrics, and page framing
- Mobile QA gate for layout regressions
What stays outside the system.
- No live wallet, trade, transfer, or fund power.
- No SAR filing, KYC approval, enforcement, or final compliance decision.
- No secrets, raw logs, customer data, or employer data.
- No guarantee that synthetic samples reflect production performance.
Named tools are useful when they clarify the system.
Some names make the record easier to understand. These anchors can support future diagrams, screenshots, or tool imagery without turning the site into a vendor pitch.
Useful visual: task card, trace row, or review checkpoint.
Useful visual: code review diff, test gate, reviewer lane, or build trace.
Useful visual: model card, cost cap, or routing decision.
Useful visual: ledger row, queue item, history row, or recovery record.
Useful visual: commit trail, issue map, release note, or validator check.
Useful visual: skill supply-chain diagram, trust boundary, or malicious-skill case note.