Production Systems

Projects

AI agents and automation systems designed and deployed to solve real operational challenges, with measurable business impact.

Featured ProjectProduction · Active

Sophia — Autonomous Multilingual AI Sales Agent

Sophia is a production AI conversational agent operating over WhatsApp Business to automate the automotive sales process from initial lead capture through appointment scheduling. She manages multilingual conversations, qualifies leads, queries live inventory, and books appointments — without human intervention for routine interactions.

The system ingests inbound leads from Meta Lead Ads via webhook, routes them through n8n orchestration workflows, and uses a LangChain agent with three registered tools: get-ucd-inventory (MongoDB query), send-stock-card (WhatsApp template dispatch), and book-an-appointment (Nylas Calendar API). Audio messages are transcribed by OpenAI Whisper before being processed.

Try Sophia on WhatsApp

Live agent — tap the button and send a message to interact with Sophia directly.
Test mode — your number will not be stored or treated as a real lead.

Project Outcomes

< 30 s

Lead response time

Down from 2–4 hours of manual follow-up

30+

Languages supported

Auto-detected via LangChain language classifier

24/7

Autonomous operation

Zero human required for routine qualification flows

Integrated systems

OpenAI, ElevenLabs, Nylas, MongoDB, WhatsApp, Meta, Sheets & more

WhatsApp conversation screenshot — Sophia in action

Full Technology Stack

Every service integrated in the Sophia system

n8n orchestrationNode.js microservicesOpenAI GPT + WhisperElevenLabs multilingual TTSWhatsApp Business Cloud APIMongoDB Atlas (chat memory + inventory)Nylas Calendar APIMeta Lead Ads webhooksGoogle Sheets API

Agent Capabilities

Multilingual Conversation (30+)

Detects language automatically on every message and responds in the customer's own language. Supports 30+ languages via OpenAI GPT with a dedicated language-classifier agent running before each LLM call.

Lead Qualification Workflows

Classifies inbound leads from Meta Lead Ads by intent and routes through structured n8n qualification workflows. Intent classification separates inventory queries, scheduling requests, and escalation paths before the main agent responds.

Real-Time Inventory Queries

Agent queries a MongoDB inventory collection using tool-use / function calling — the LLM generates the query, executes it, and returns a formatted vehicle card (image, price, specs, action buttons) via WhatsApp template messages.

Appointment Booking via Nylas

Books test drives and appointments by calling the Nylas Calendar API sub-workflow. The agent extracts appointment time (ISO 8601), client name, email, and salesperson preference via $fromAI() and creates a calendar event with all participant metadata.

Voice Audio Greetings (ElevenLabs)

On first contact, sends a personalized voice audio greeting in the customer's language using ElevenLabs eleven_multilingual_v2. Uses language-specific voice models and delivers via WhatsApp audio message. Audio session is stored in MongoDB chat memory.

Persistent Session Memory

Each conversation is stored in MongoDB (dbSophia / chatMemory collection), keyed by WhatsApp ID. The agent retrieves the last 10 messages as context on every turn — maintaining continuity across disconnected sessions.

Additional Automation Systems

Beyond Sophia, a suite of supporting automation systems was built across the same dealership operation — each solving a distinct operational problem.

Financial Document AI

Google Cloud

Implemented Google Cloud Document AI processors to automate data extraction from automotive funding forms (credit applications, lease agreements). Replaced manual data entry with structured field extraction, reducing errors by 90%.

Pricing Audit Agents

Internal Tool

Agents that compare manual internal pricing records against live dealership website data on a scheduled basis. Detects discrepancies and alerts management of inconsistencies automatically — no manual audit required.

Meta Ad Catalog Automation

Marketing

Automated dynamic retargeting campaign catalog generation via CSV manipulation and Meta Ads API. Inventory changes automatically propagate to active ad campaigns without manual intervention.

Predecessor System

NewKeys.Link — Node.js Lead Processing Engine

Before migrating to n8n, the lead processing pipeline was built entirely in Node.js — custom webhook controllers for Meta Lead Ads, real-time SMS validation via Twilio, modular route/service architecture, and asynchronous event processing. This system was the technical foundation that later evolved into the Sophia AI ecosystem. It proved the architecture patterns that n8n now orchestrates at scale.

Real-time lead processing logs: webhook reception, WhatsApp template dispatch, and async event handlers.

Coming Soon

Interactive Demos

Future versions of this portfolio will embed interactive demos directly in the browser. For now, Sophia is accessible live via WhatsApp.

Get in Touch