PertoDaqui AI & Messaging • Technical Case Study

AI Assistant "O Daqui"

AI portal assistant and lead qualification engine

› Screenshots

3 planned
Sanitized preview of portal conversational flow showing dynamic vector knowledge retrieval.

Sanitized preview of portal conversational flow showing dynamic vector knowledge retrieval.

PREVIEW
#01

Sanitized assistant conversation.

· coming soon ·
#02

Intent/cache/retrieval flow.

· coming soon ·

› Technical Stack

LangGraph.ts Intent Router Vector Database FAQ Cache

› Architecture Pipeline

Message →Intent Router →Cache / Vector DB / LLM →Lead Extraction →Budget CTA

› The Problem

The public portal assistant must answer quickly, qualify leads naturally and avoid paying for an LLM call for every simple FAQ.

› What I Built

I built O Daqui as the front-door assistant for PertoDaqui: local intent routing, FAQ cache, vector retrieval, LangGraph state, lead extraction and CTA triggering.

› Core Capabilities

  • Local intent router for common questions.
  • FAQ cache and vector retrieval before generation.
  • LangGraph state for multi-turn lead qualification.
  • Incremental extraction of contact, business type and location.
  • Budget CTA when enough lead data is present.

› Technical Decisions

01

Route locally first to reduce cost and latency.

02

Use retrieval only when the answer needs platform knowledge.

03

Call the LLM after cheaper routes fail.

04

Collect lead data conversationally instead of forcing a long form.

› What This Demonstrates

Cost-aware conversational AI, retrieval, caching, stateful lead qualification and practical conversion design.

› Limitations & Next Steps

  • Add voice input once the typed flow is stable.
  • Tune extraction latency on smaller deployments.

› Project Links