Credit Union Operations & AI Platform

01The problem

Credit unions manage members, savings and share accounts, loans, front-desk cash, and accounting, usually across disconnected tools and spreadsheets.

A branch manager can't answer basic questions in one place: What happened today? How much money moved? Which loans need attention? Which approvals are waiting? Loan risk is judged case by case, with no early warning.

The goal: one web platform that many credit unions can use, each with its own branding, users, members, and dashboards, with AI-assisted risk insight as a premium tier.

02My role

• Owned the program end-to-end: product structure, written build plan, module breakdown, and delivery sequencing.
• Translated credit union operations into requirements: roles, permissions, approval rules, and daily workflows for every user type.
• Designed the management reporting layer, defining the questions each dashboard must answer for managers and the board.
• Ran QA and a full platform audit, and tracked every finding to verified closure in a living status tracker.
• Built hands-on with AI-assisted development, coordinating the product, data, ML, and infrastructure workstreams.

03What was delivered

• Multi-tenant SaaS on ERPNext/Frappe with two plan tiers: Basic (core operations) and Standard (adds AI/ML risk insight).
• Role-based workflows for Front Desk/Teller, CRM & Loan Officer, Branch Manager, Accountant, and a read-only Board view.
• Operational modules: member onboarding & KYC, savings & share accounts, front desk, loan lifecycle (application → approval → disbursement → repayment), accounting, HR, internal projects, and assets.
• Management dashboards: Command Center, Decision Center, Collections Workbench, Loan Portfolio, Deposits & Shares, Member and CRM Intelligence, Compliance & Service, and more.
• AI layer: loan default-risk and repayment-delay models, collection prioritization, and a "Talk to Data" assistant that answers management questions in plain language.
• Controls: a large-withdrawal approval gate based on a member's cumulative same-day total, enforced account holds, and branch-scoped data access.

04How the program was run

• Plan first. A written build plan defined product tiers, modules, user roles, and what each role can and cannot do before build started.
• Walk the real flows. 13 end-to-end user flows (members, accounts, front desk, loans, operations, reports, scheduled jobs, security) were documented and walked as each role would use them.
• Audit and track to closure. A structured audit covered security, fresh-install reliability, data honesty, correctness, and process health. Every finding was logged with its status, evidence, and the verification performed.
• Only real numbers on dashboards. Placeholder and estimated figures were replaced with values from real daily snapshots. Where no real data exists, the dashboard shows a dash instead of a plausible-looking guess.
• Verify, don't assume. A fix counted as done only after it was checked live as the real user role.

05Results

The dashboards run on the Visual Intelligence Library, a parallel workstream of the same program.