Back to portfolio
Project dossierSIGNAL-ATLAS.CASE
Live2026Featured

CheckWise - REPSE Compliance Platform

A REPSE-compliance SaaS where vendors submit recurring evidence, LegalShelf reviewers adjudicate, and clients see portfolio-wide risk.

A REPSE-compliance platform for Mexico. Vendors upload monthly, bimonthly, four-monthly, and annual evidence (SAT, IMSS, INFONAVIT, acuses, corporate file); LegalShelf reviewers approve, reject, or request clarifications; and client companies read a portfolio-wide risk view.

CheckWise reviewer decision queue screenshot

Problem

What this project solves

REPSE compliance in Mexico requires vendors to submit recurring, institution-specific evidence on different cycles. Tracking it manually across many vendors is error-prone and opaque, and clients have no clear portfolio-wide view of their exposure.

Solution

How I approached it

CheckWise centralizes vendor evidence intake, the reviewer adjudication workflow (approve, reject, request clarification), and a portfolio-wide client risk view, with role-based access and an AI-assisted reports center that stays explainable.

Architecture

System structure

01

Next.js 15 / React 19 frontend with a locked design system applied across admin, client, and portal surfaces.

02

FastAPI backend separates auth, RBAC, vendor expedientes, reviewer decisions, and reports.

03

PostgreSQL via SQLAlchemy and Alembic models vendors, evidence, institutions, cycles, and audit logs.

04

Reports layer adds an AI planner and streaming generator with per-block regenerate and explain, guarded by an AI-safety test suite.

Decisions

Tradeoffs and outcomes

01

Decision 1

Tradeoff: Modeling REPSE's recurring, institution-specific evidence cycles.

Outcome: Finish wiring the remaining dashboards from mock adapters to live backend endpoints.

02

Decision 2

Tradeoff: Holding one visual direction across roughly twenty admin, client, and portal surfaces.

Outcome: Move storage from the local filesystem to an S3-compatible production backend.

03

Decision 3

Tradeoff: Keeping AI report generation safe and explainable with a dedicated test suite.

Outcome: Replace the provider portal's opaque workspace token with full JWT auth.

Proof

Evidence and impact

01

A real client SaaS for LegalShelf: a full-stack build spanning a FastAPI backend, a Next.js 15 frontend with a locked design system across ~20 surfaces, role-based auth, and an AI-assisted reports flagship backed by 320+ backend tests including an AI-safety suite.

02

Reports flagship: AI planner + streaming generator + embedded copilot

03

320+ backend tests including an AI-safety suite

04

Role-aware reviewer queue with JWT auth and RBAC

05

REPSE calendar (institution x month) and expediente gates

Roadmap

Next iteration

01

Finish wiring the remaining dashboards from mock adapters to live backend endpoints.

02

Move storage from the local filesystem to an S3-compatible production backend.

03

Replace the provider portal's opaque workspace token with full JWT auth.