Founder-led program · AIBYTERS · Financial services SaaS
Credit Union Operations & AI Platform
How I took a multi-tenant credit union platform from a written build plan to an audited, hardened product, with role-based workflows, approval controls, management dashboards, and ML risk scoring.
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.
Let's find the bottleneck in your operations.
A 30-minute call is enough to map where work gets stuck and agree on a practical first step.