Ayoub Taoufik

Project 02 · Case study

HelloLeo — AI product platform

Human-facing tools, business-system integrations, testing, guardrails, and delivery for AI-enabled business applications.

Full-stack engineer · co-built with the HelloLeo team

At HelloLeo, the work was not limited to putting an LLM behind an interface. I co-built with the team across the product system that makes AI features useful in business workflows: the human-facing tools, connections to business systems, testing, guardrails, and the path from feature work to delivery.

01 · Product context

AI capability had to become usable software

An AI feature becomes useful in business software only when people can use it through a product surface, when it can work with the right tools and connected systems, and when change can be tested and delivered with care.

That made the work a full product-system problem rather than an isolated model interaction.

02 · My contribution

The systems around the model

I built the human-facing layer over tool connections and integrations, including MCP servers and connectors to business systems and ERPs. I also built integrations with multiple LLM providers.

For product quality and delivery, I built testing pipelines including A/B testing of system prompts, implemented guardrails for AI-enabled workflows, and worked across deployment pipelines.

  1. 01 · Tools

    Connections that belong in the product

    Tool and business-system connections were treated as part of the user-facing product path, not as separate technical capabilities.

  2. 02 · Change

    Evaluation as feature work

    Testing pipelines and system-prompt A/B testing made behaviour something that could be worked on deliberately as the product changed.

  3. 03 · Delivery

    A route from work to release

    Guardrails and deployment work remained connected to feature delivery rather than being left as an afterthought.

03 · Working boundary

A clear AI engineering boundary

This is AI product engineering around models: product integration, tools, testing, guardrails, and delivery. It is not a claim of model training, ML research, or ownership of the platform’s overall architecture.

Evidence treatment · contribution map

The parts of the product I worked on

Scope is deliberately grouped by my ownership boundary. It is not a claim about the whole system.

01

Built

  • The human-facing tool and integration layer.
  • MCP servers and connectors to business systems and ERPs.
  • Integrations with multiple LLM providers.
02

Built for product quality

  • Testing pipelines, including A/B testing of system prompts.
  • Guardrails for AI-enabled workflows.
03

Worked across

  • Deployment pipelines for the product.