simone@renor:~$ cat ~/progetti/runai.md

RunAI: AI infrastructure and APIs

RunAI in Simone Renzi’s portfolio: OpenAI-compatible APIs and AI infrastructure.

The workflow at a glance

  1. Question
  2. Server + profile
  3. RunAI API
  4. Streaming answer

Functional diagram of the example described below.

The problem: connecting an application to models

An AI application has distinct responsibilities: its interface, the context it supplies and the inference system. RunAI is an AI infrastructure platform with OpenAI-compatible APIs. Applications can use a common interface while keeping model execution separate. API compatibility does not imply identical capabilities or quality across models.

Simone Renzi’s role

Simone Renzi handled conception, architecture, development and platform management. The project combines management of model-serving infrastructure with the software that exposes it to applications. One concrete integration is available on this website: “Ask Simone’s CV”.

The CV module’s workflow

The page collects a question. A server component adds the curated professional profile and recent conversation exchanges, then sends the request to RunAI’s API. The answer streams back to the browser. API credentials remain on the server and are not part of the public page.

Verifiable integration choices

The component limits questions to 240 characters and includes at most six previous exchanges. Context includes the profile and credentials published on this site. The instructions ask the model to use only information in the profile and acknowledge missing facts. These constraints narrow the task; they do not guarantee every answer is correct.

What you can observe

Open “Ask Simone’s CV” in the engineering profile and ask about Simone Renzi’s background. The example shows a public website connected to an inference service, with progressive output and application-controlled context. The CV remains the reference for checking factual information.

Limits and evaluation

This case study documents an integration, not a benchmark of capacity, financial savings or accuracy. Performance and requirements depend on the model, context and concurrent requests. A different deployment needs representative examples, acceptance criteria and review of the complete data path. The linked guides explain how to approach that comparison.

Explore further