Drisatech/Drisa_AI-Multilingual-Support-Agent
Domain:
natural language processingRecord type:
software
Creator:
Dri
Host:
Drisa_AI is an autonomous voice agent for sales and support. Powered by Gemini, it answers calls in English, Hausa, Igbo, Yoruba, and Pidgin. It queries live website catalogs and sends automated product follow-ups via WhatsApp or Email.
# Drisa_AI Agent
A professional multilingual AI Voice Support Agent built with React, Express, and Gemini 2.5 Flash.
## Architecture
### đź“Š System Flow (Mermaid)
```mermaid
graph TD
subgraph Client_Layer [Client Layer]
A[User Web App - React]
B[Admin Dashboard - React]
C[Phone / Telephony - Twilio]
end
subgraph Backend_Core [Backend Core - Node.js/Express]
D[Session Manager]
E[Audio Transcoder]
F[Tool Orchestrator]
G[RAG Engine]
end
subgraph AI_Intelligence [AI Intelligence]
H[Gemini 2.5 Flash Live API]
end
subgraph Data_Persistence [Data Persistence]
I[(Google Firestore)]
M[Admin Request Flow]
end
subgraph External_Integrations [External Integrations]
J[Google Calendar API]
K[Meta WhatsApp API]
L[Nodemailer / SMTP]
end
A |WebSocket| D
C |Media Stream| D
B |REST API| D
D E
D F
F G
D |Live Stream| H
F I
F --> J
F --> K
F --> L
B -->|Admin Request| M
M -->|Email Alert| L
```
### 🔄 Workflow Overview
1. **Multimodal Ingress**: Users engage via a high-fidelity React web interface or traditional telephony (Twilio), establishing a low-latency WebSocket or Media Stream connection.
2. **Real-time Stream Orchestration**: The Node.js/Express backend acts as a high-performance proxy, performing real-time audio transcoding (mu-law ↔ PCM) and managing stateful sessions.
3. **Gemini 2.5 Flash Intelligence**: The core AI engine utilizes the **Gemini 2.5 Flash Multimodal Live API** to process raw audio streams directly, enabling native understanding of tone, emotion, and multilingual nuances.
4. **Autonomous Tool Execution**: Based on user intent, the agent autonomously invokes specialized tools to query the Firestore-backed product catalog, schedule appointments via Google Calendar OAuth, or dispatch follow-ups via Meta WhatsApp and SMTP.
5. **Persistent Memory & Analytics**: Every interaction is captured in **Google Firestore**, providing a rich dataset for lead management, conversation history, and business intelligence.
## 🌟 Key Fe …