Paul Dosser
FREN
Back to home

Case study - Client work

PMS - An AI-native project management platform

A custom PMS built for a B2B services client. Twelve AI features, sixty-plus granular permissions, and a role-based governance model that keeps the admin in full control of the AI layer.

AI features
12
Granular permissions
60+
Feature domains
10
Languages (FR/EN/ES)
3

The brief

The client needed a project management tool that actually matched their workflow. Off-the-shelf tools couldn't model their billable/non-billable statuses, their role hierarchy (admin, PM, dev, client), their maintenance ticketing flow, or their client-facing reporting obligations.

AI had to be woven into daily operations rather than bolted on, and the admin team needed to stay in full control of who gets access to what.

The challenge

Before PMS, project managers were writing daily and weekly reports by hand. Task scoping was inconsistent, so estimates drifted. Client feedback arrived through scattered channels and was hard to triage.

Because every client and every role had slightly different access needs, the permission model had to be granular without becoming unusable.

The solution

PMS is a full-stack project management platform with a governed AI layer sitting on top of a fine-grained permission system.

Every AI feature can be independently toggled per role by the admin — nothing is forced on anyone. The permission matrix is backed by Postgres RLS, not just UI guards, so access is enforced at the data layer.

Governed AI layer

Each AI action is gated by a per-role feature flag stored in app_settings. Admins see a matrix and flip switches. Non-admins only see AI features their role is authorized for.

  • Predict Hours

    Task

    AI estimates hours needed for a task based on its description.

  • Generate Todolist

    Task

    Auto-generate a checklist of subtasks from a task description.

  • Enhance Description

    Task

    Improve task descriptions to be clearer and more actionable.

  • Summarize Chat

    Chat

    Summarize chat conversations into key points and action items.

  • Draft Invoice

    Invoice

    Suggest invoice line items from project/task context.

  • Project Health

    Project

    Analyze project metrics and provide a health assessment.

  • Classify Feedback

    Feedback

    Automatically categorize and prioritize client feedback.

  • Generate Report

    Project

    Generate summary or detailed reports from project data.

  • Client Report

    Company

    Generate a client-facing progress report from real activity.

  • Polish Message

    Chat

    Rewrite chat messages for clarity and professionalism before sending.

  • Translate Resource

    Learning

    Translate learning resources into other languages.

  • Permission Advisor

    Admin

    Audit role permissions with security suggestions (unused grants, over-broad roles, friction points).

Role permissions

60+ permissions across 10 feature domains, surfaced through an admin matrix and backed by Postgres RLS. A dedicated AI advisor audits the matrix and flags unused grants, over-broad roles, and friction points with confidence scores.

Domains covered

  • Projects
  • Phases
  • Tasks
  • Feedbacks
  • Maintenance
  • Teams
  • Medals
  • Invoices
  • Settings
  • Users

Tech stack

  • Next.js 16 (App Router, Server Components)
  • React 19
  • Supabase (Postgres, Auth, RLS)
  • Anthropic Claude SDK
  • next-intl (FR/EN/ES)
  • Tailwind CSS 4
  • Zod, Playwright

Outcomes

Live and actively used by the client's team. Daily and weekly reports are generated from real activity rather than written by hand.

Feedback triage is automated. The permission system evolves with the business without manual audits thanks to the AI advisor.