Raposa Aval — Human Approval for AI Agents

Raposa Aval is an independent DC ESCRYPT engineering project for adding a human decision point to AI-agent workflows.

An agent may be able to call a tool, but that does not mean every action should execute automatically. Sensitive or irreversible actions need an approval boundary outside the model, with a clear record of what was proposed and what a person decided.

I worked as Founder and AI Systems Engineer: I defined the approval model, designed the integration boundary, implemented the Python/MCP component and supporting API surfaces, and documented the workflow for practical integration.

Python service and MCP integration

REST API connectivity

Human-in-the-loop approval flow

Controlled tool execution

Auditable decision records

Public web and documentation surfaces

Separates an agent's proposed action from authorization to execute it

Records explicit approval, rejection, or expiry decisions

Keeps the approval decision outside the agent's own authority

Provides a controlled path for integrating approval into agent tools and workflows

Documents the system for use from MCP-compatible clients

The project has a public product surface at https://raposa.group and a published Python MCP component, as reflected in the Upwork portfolio record and the project materials. The case study describes the implemented and published surfaces; it does not claim client revenue, contract history, or performance metrics.

The cover image is a screenshot of the public Raposa interface, not a speculative mockup.