ai/ml

2024-Present

Case Study

Fyncall

AI-powered customer service platform with multi-agent systems, unified messaging, and intelligent automation.

Live Website

Key Metrics

100K+

Lines of Code

150+

API Endpoints

35+

AI Tools Built

15+

OAuth Integrations

Overview

Fyncall is a comprehensive AI-powered customer service platform that I co-founded and built as CTO. The platform revolutionizes how businesses handle customer interactions through intelligent automation, multi-agent AI systems, and unified messaging across all channels. From architecture to deployment, I built the entire technical infrastructure from the ground up.

My Role

Co-Founder & CTO

As CTO, I'm responsible for the entire technical vision, architecture, and implementation. I designed and built all backend systems, AI pipelines, integrations, and infrastructure. I also lead technical decisions, manage the development roadmap, and ensure the platform meets enterprise-grade reliability standards.

Challenges

Multi-Channel Message Unification

Customers expect seamless conversations across WhatsApp, Instagram, Email, SMS, and web chat. Each platform has different APIs, rate limits, and message formats that needed to be normalized into a single conversation thread.

Real-Time AI Response Generation

AI responses needed to be fast (<2 seconds), contextually aware of the entire conversation history, and capable of accessing business-specific knowledge bases without hallucination.

Complex Multi-Agent Orchestration

Different customer queries require different specialized agents (sales, support, billing). Routing queries to the right agent while maintaining conversation context was a significant architectural challenge.

Enterprise Integration Requirements

Businesses need AI that can actually take actions - process refunds in Stripe, check orders in Shopify, update records in their CRM. Building secure, reliable integrations at scale was critical.

Solutions

Unified Message Bus Architecture

Built a message normalization layer that converts all platform-specific formats into a unified schema. WebSocket connections maintain <100ms latency for real-time updates across all channels.

RAG-Powered Knowledge System

Implemented a Retrieval-Augmented Generation system with semantic search that achieves sub-second retrieval from knowledge bases. This ensures AI responses are grounded in actual business data.

LangGraph Supervisor Pattern

Designed a multi-agent system using LangGraph where a Supervisor agent routes conversations to specialized agents (Sales, Support, Billing) based on intent classification and conversation context.

MCP Protocol for Tool Integration

Implemented Model Context Protocol (MCP) for standardized tool calling. This allows AI agents to securely interact with external services like Shopify, Stripe, and custom APIs.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        CLIENT LAYER                             │
├─────────────────────────────────────────────────────────────────┤
│  WhatsApp  │  Instagram  │  Email  │  SMS  │  Web Widget       │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                    MESSAGE GATEWAY                              │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐         │
│  │ Webhook      │  │ WebSocket    │  │ Message      │         │
│  │ Handlers     │  │ Server       │  │ Queue        │         │
│  └──────────────┘  └──────────────┘  └──────────────┘         │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                     AI ENGINE                                   │
│  ┌──────────────────────────────────────────────────────────┐  │
│  │                   SUPERVISOR AGENT                        │  │
│  │  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐     │  │
│  │  │ Sales   │  │ Support │  │ Billing │  │ Custom  │     │  │
│  │  │ Agent   │  │ Agent   │  │ Agent   │  │ Agents  │     │  │
│  │  └─────────┘  └─────────┘  └─────────┘  └─────────┘     │  │
│  └──────────────────────────────────────────────────────────┘  │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐         │
│  │ RAG Engine   │  │ Tool         │  │ Memory       │         │
│  │ (Knowledge)  │  │ Executor     │  │ Manager      │         │
│  └──────────────┘  └──────────────┘  └──────────────┘         │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                   INTEGRATION LAYER (MCP)                       │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐           │
│  │ Shopify │  │ Stripe  │  │ Gmail   │  │ Twilio  │           │
│  └─────────┘  └─────────┘  └─────────┘  └─────────┘           │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                    DATA LAYER                                   │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐         │
│  │ PostgreSQL   │  │ Redis        │  │ Vector DB    │         │
│  │ (Primary)    │  │ (Cache/Queue)│  │ (Embeddings) │         │
│  └──────────────┘  └──────────────┘  └──────────────┘         │
└─────────────────────────────────────────────────────────────────┘

Technical Highlights

Multi-Agent LangGraph System

Built a sophisticated agent orchestration system using LangGraph with a Supervisor pattern. The system dynamically routes conversations to specialized agents based on intent classification, maintaining full conversation context across agent transitions.

Sub-Second RAG Retrieval

Implemented a high-performance RAG system using vector embeddings and semantic search. Knowledge base queries return relevant context in <500ms, enabling real-time AI responses grounded in business data.

35+ AI Tools with MCP

Developed 35+ intelligent tools that AI agents can use to interact with external services. Using the Model Context Protocol ensures secure, standardized tool execution with proper error handling.

Real-Time WebSocket Infrastructure

Built a WebSocket infrastructure that handles thousands of concurrent connections with <100ms message latency. This enables real-time chat experiences across all messaging channels.

Complete OAuth Integration Suite

Implemented OAuth 2.0 flows for 15+ services including Shopify, Instagram, Gmail, and Stripe. The system handles token refresh, scope management, and secure credential storage.

Full Observability Stack

Deployed a complete monitoring solution with Grafana dashboards, Prometheus metrics, and Loki log aggregation. This provides real-time visibility into system health, AI performance, and user interactions.

Results & Impact

100K+

Lines of Production Code