Ayoub Taoufik

Case 03

HelloLeo: AI product platform

AI features built around each company’s ERP (business software for running core operations), with custom fields read first and production changes approved by a person.

Live product · 2024–2026

Full-stack engineer

HelloLeo builds AI features around each company’s ERP setup. I worked across every layer of the product, from its platform core and deployment pipelines to the tools people use, MCP servers, ERP and business-system connectors, integrations with multiple LLM providers, system-prompt testing, and workflow guardrails.

Visit HelloLeo
The problem
AI features had to read each company’s own ERP modules and fields, while production changes and outbound work required a person’s approval.
What I built
I worked across the platform core and deployment pipelines, user-facing tools, MCP servers, ERP connectors, multiple LLM providers, system-prompt A/B tests, and workflow guardrails.
What changed
The public product reads ERP modules, fields, and relationships before generating work. It keeps staging isolated, requires approval for production changes, and asks before an agent sends work out.
Fig. 03.1Every company sets up its ERP its own way, so generic AI templates don’t fit.HelloLeo reads the company’s own modules and fields first, then builds features, automations and reports on them.

The ERP card shows an example custom Fleet module with its own fields. The report repeats those same columns and marks the sample rows as examples. A model sits beside the workflow, which ends at staging, a person’s approval, and live.

Your ERP, read first

e.g. custom moduleFleet

PlateDriverNext service

Its own modules and fields.

Under the hood: MCP servers · ERP connectors
Built on your fields

Vans due for service

PlateDriverNext service
VAN-12Alex12 Oct
VAN-28Sam14 Oct
e.g. fleet report
FeaturesAutomationsReports

Uses those exact fields, not a generic template.

Staging → A person approves → LiveDeployment · testing pipelines · approval gate
Fig. 03.2 · Film · 0:10
HelloLeo reads a company’s own ERP fields first, here a custom Fleet module, then builds a report that uses those exact fields.
Transcript

This is an illustrative flow, not a recording: HelloLeo first reads the company’s own ERP setup, a custom Fleet module with Plate, Driver and Next service fields. A report, Vans due for service, is then built on those same fields and fills with two vans. The fleet data are examples.

AI features built for each company’s ERP

HelloLeo builds AI features inside ERP systems, the business software companies use to run core operations. Each company can have its own modules, fields, and relationships, so the product reads that setup before generating anything.

The public product flow is connect, prompt, customize, and publish.

The systems behind the AI features

I worked across the platform core and its deployment pipelines, the user-facing tools, MCP servers, connectors to ERPs and other business systems, and integrations with multiple LLM providers.

I built pipelines for system-prompt A/B testing, added guardrails to AI workflows, and built the deployment pipelines.

  1. Tools

    ERP and business-system connections

    I built the tool connections into the product experience people use.

  2. Quality

    System-prompt testing

    Testing pipelines checked system prompts as the product changed.

  3. Delivery

    Deployment and release

    I built workflow guardrails and deployment pipelines as part of product delivery.

Staging first, production on approval

Work goes to an isolated staging environment first. A person must approve production changes, and agents ask before sending work out.

The public product follows this release path through the platform core and deployment pipelines.

My role in AI product engineering

My work covered product integration, tools, prompt testing, guardrails, and deployment. I did not work on model training or ML research.

Early prototypes included a parser for definitions and references in source code, and an app starter. These were explorations, not part of the shipped product.

Scope

What I built

Built

01
  • User-facing tools and integrations.
  • MCP servers and connectors to ERPs and other business systems.
  • Integrations with multiple LLM providers.

Prompt testing and workflow guardrails

02
  • Pipelines for testing system prompts.
  • Guardrails for AI workflows.

Platform and deployment

03
  • The platform core and its deployment pipelines.

Stack

  • MCP
  • LLM-provider APIs
  • ERP connectors