Agnicio SSP

A financial operator lost hours each week to spreadsheets — risk scores nobody could explain, and an audit trail that demanded an answer for why a deal got approved. Legacy tooling gave the number but hid the reasoning.

• Discovery of the process and written spec before the first invoice
• Product design and UI system in Figma, validated against real data
• Full-stack build on Next.js + Postgres + Supabase Auth
• LLM steps integrated to explain scoring decisions — not to make them
• Deploy, monitoring, and a handover packet for their internal team

We watched their team work for a week before writing code. We modelled the scoring pipeline as a composition of explainable steps, not a black box — every number links back to the supporting data and the input parameters. The LLM layer summarises the reasoning in human language but never owns the decision. The audit log shows exactly who changed what and when.

• Next.js
• TypeScript
• Postgres
• Supabase Auth
• Tailwind
• Vercel

1. 01Onboarding a new operator dropped from weeks to days
2. 02100% of scoring decisions now carry an audit-ready explanation — no manual work
3. 03SSO + role-based access covers the split between analyst, audit, and management
4. 04Running in production on Vercel with monitoring and weekly backups