Personit Hub - Case Study

Case study React Supabase AI-powered Web App SaaS

We built an AI-first system for ourselves — and it gave us back 200 hours a month

90%

of tasks are created automatically from meeting transcripts and email

95%

of status changes happen without a human touching them

200 h

recovered every month from operational project work

About Personit Hub

Personit's internal operating system — sales, delivery and the team's daily work in one place. Projects here largely run themselves: tasks appear straight from what was agreed, statuses change in the background, and the AI agent stops where a person has to decide. The whole team uses it every day, across the dozens of projects we run in parallel.

About Personit

At Personit, we help startups and scaling businesses move from idea to real product — up to 5x faster than traditional dev teams. We specialize in low-code platforms and AI-powered development. Hub is the only case study in our portfolio where the client was us: the methods we use for our clients were tested on our own delivery first.

Challenge

When you run dozens of projects at once, traditional project management becomes a slog. Someone has to keep tasks in sync, chase deadlines and stand between the client and the team. None of that ships product — it is work about work. We needed a system that would:

Stop what leaks — an arrangement made with a client cannot live in the memory of one person who happens to be on holiday
Turn what was agreed in a meeting into assigned tasks, with no manual note-taking
Answer everyone's "what am I doing today" without a status meeting
Separate the sales funnel from the delivery board instead of forcing both into one status field
Show project profitability as it changes, not in a month-end report
Let an AI agent act on its own where it is safe — and stop it where a person has to decide

4 weeks

Key Functionalities

An AI agent in every meeting — decisions become tasks, not notes

Tasks created automatically and assigned to named people

Statuses updated in the background — from commits and coding agents

Decision inbox: the agent proposes, a person approves in one click

Profitability recalculated from logged hours, with no report to run

Sales funnel and delivery board, independent, on a single project

Why low-code + AI?

A system where an AI agent writes to the database alongside people had to ship fast — but above all it needed hard limits on what the agent may do on its own. That is why we built the backend on Supabase.

React & Supabase allowed us to:

Building the interface with ready-made React components instead of coding UI from scratch
Supabase as 'Backend-as-a-Service' — accounts, authentication and API access right out of the box
Postgres with Row-Level Security as the home of the agent's boundaries: it cannot approve itself or read margins, because no access policy grants it
Realtime for boards that update in front of the whole team
Edge Functions for AI model calls and integrations, with no servers to maintain
Security and scalability 'out-of-the-box', saving weeks of development work

The result: a system where the agent's autonomy is a dial in the interface, and the database — not the prompt — is what holds the line.

Low-Code Speed, Custom Code Quality

Rapid launch thanks to ready-made backend modules

Agent boundaries enforced in the database, not in a prompt

Every change signed by its author: a person, an agent or a commit

AI models running on Edge Functions — no servers to maintain

Our team has delivered over 100+ production-grade apps using technologies such as React, FlutterFlow, Bubble, Supabase, Firebase and AI-enhanced tooling. That experience allows us to go fast without cutting corners – a key differentiator in the low-code space.

The Results

90% of tasks in the system are created without a human

Nothing falls through with a client — every arrangement lands in the system and gets an owner

The whole team knows what to do every day — with no status meeting

95% of status changes happen on their own; the remaining 5% is left to people on purpose

Every project's profitability is visible instantly, with no report and no spreadsheet

Dozens of projects running in parallel, with no operational work around them

Our Approach

Discovery

Design

Development

QA / Iteration

Who built it

Mikołaj

Tech Lead

Mikołaj is a Tech Lead who combines deep technical expertise with a natural talent for leading teams. His superpowers? Rapid development in its purest form – he builds scalable applications at lightning speed, never compromising on best practices and code quality. He's equally skilled at designing system architecture and inspiring teams to push beyond their limits. Thanks to the perfect balance between technical vision and operational excellence, every project under his leadership gains solid foundations and incredible delivery pace.

Anna

Web developer / Tech Educator

Anna is a skilled Bubble, React developer and technology educator who brings creativity and technical expertise to every project. She specializes in building complex web applications using both low-code tools and traditional frameworks. As a mentor and educator, she enjoys sharing her knowledge, supporting the growth of other professionals.

Where our experience made a difference

The right question is not "what else can we automate" but "what should we never automate" — that 5% left to people is a design decision, not a backlog item.

An agent is only as good as its access to context. It stopped guessing the moment we connected the calendar, the transcripts and the inbox.

When half the writes come from machines, "who did this?" becomes an interface question — which is why every history row carries its author's mark.

A proposal must never look like a fact. A dashed border around a proposed status does more for trust than any amount of documentation.