PertoDaqui AI & Messaging • Technical Case Study

Intelligence

Configurable AI control plane for assistants, APIs, BYOK and hybrid model routing

› Screenshots

3 planned
Control dashboard showing real-time configuration of assistant prompts and model routing.

Control dashboard showing real-time configuration of assistant prompts and model routing.

PREVIEW
#01

Assistant configuration dashboard.

· coming soon ·
#02

BYOK/settings view with secrets hidden.

· coming soon ·

› Technical Stack

AI Control Plane BYOK Hybrid LLMs Guardrails

› Architecture Pipeline

Dashboard Config →REST APIs →Assistant Runtime →Hybrid LLM Router →Fallback Logic →Business Workflows

› The Problem

A useful AI layer for local businesses cannot be a single chatbot prompt. It needs client settings, sector prompts, model choices, fallbacks, API-key ownership and safety controls.

› What I Built

I built PertoDaqui Inteligencia as the AI control plane: dashboard-managed assistant configuration, sector behaviour, BYOK settings, runtime APIs, model routing and fallback/guardrail rules.

› Core Capabilities

  • Assistant configuration per client and business sector.
  • BYOK support for client-owned Groq keys and platform fallbacks.
  • Prompt, welcome-message and runtime behaviour management.
  • Hybrid routing across cloud and local/controlled models.
  • Guardrails for injection, spam, message size and unavailable APIs.

› Technical Decisions

01

Separate configuration from runtime so behaviour can change without redeploying.

02

Make API-key ownership explicit instead of hiding all cost behind the platform.

03

Run guardrails before model calls to reduce unsafe or wasteful inference.

04

Keep static fallback responses available for degraded states.

› What This Demonstrates

Applied AI implementation beyond prompting: control-plane design, BYOK, fallback engineering, guardrails, model routing and operational AI governance.

› Limitations & Next Steps

  • Broaden local-model support for high-sovereignty clients.
  • Improve retry and backoff behaviour around inference rate limits.

› Project Links