WEB APPLICATION

Drishti Marketing OS — Enterprise WhatsApp AI Agent Platform

Enterprise WhatsApp Business inbox replacing third-party BSPs with a custom Meta Cloud API integration, multi-provider AI agent (Claude/GPT-4o/Gemini), RAG knowledge base, and real-time Zoho CRM data — achieving 89.7% AI self-service rate and 96% reduction in lead response time.

View Live Source Code

Drishti Marketing OS — Enterprise WhatsApp AI Agent Platform

Client

MaxFate Private Limited

Timeline

May 2026 – Jun 2026

My Role

Sole Developer & Technical Lead

Project Overview

Drishti Marketing OS is an enterprise WhatsApp Business inbox and AI agent platform built for MaxFate, an astrology services company. It replaces per-agent monthly subscriptions to third-party WhatsApp BSPs (WATI/AISensy) with a direct Meta Cloud API integration, eliminating BSP markup and providing full control over the messaging pipeline. The platform features a multi-provider AI switchboard (Claude 3.5 Sonnet, GPT-4o, Gemini 1.5 Pro) with an Abstract Factory pattern for runtime provider swapping, a Qdrant-powered RAG knowledge base for brand-specific context, real-time Zoho CRM contact and invoice data surfaced inside every chat thread, BullMQ queue-first webhook architecture meeting Meta's strict 2-second SLA, and a Next.js 3-pane operator console. Built in 3 weeks from zero — 163 commits, sole developer, zero deployment downtime.

Key Features

Feature Groups

Features

1 / 0

Key Capabilities

Direct Meta WhatsApp Cloud API integration — replaced WATI/AISensy BSP entirely

#01

Multi-provider AI switchboard (Claude 3.5 Sonnet, GPT-4o, Gemini 1.5 Pro) with Abstract Factory pattern

#02

Qdrant vector DB RAG pipeline for brand knowledge retrieval

#03

Real-time Zoho CRM contact and Zoho Books invoice data in every chat thread

#04

BullMQ queue-first webhook architecture meeting Meta's 2-second SLA

#05

Agentic tool-calling pipeline (lookupCustomer, getReportStatus, verifyPayment, shareReportLink, escalateToHuman)

#06

Next.js 3-pane operator console with Socket.io real-time sync

#07

TOTP 2FA authentication with staged login and QR enrollment

#08

Redis caching for Zoho API rate-limit protection (30-min TTL)

#09

3-stage Docker build with Bun runtime for minimal container footprint

#10

GitHub Actions FIFO deployment lock preventing concurrent deploy races

#11

MCP (Model Context Protocol) analytics endpoint for external AI agents

#12

Drop-in /api/external/send-template API replacing WATI calls in existing Zoho CRM workflows

#13

Case Study

Problem & Approach

The Challenge

Problem Statement

MaxFate was paying per-agent monthly subscriptions to third-party WhatsApp BSPs (WATI/AISensy) and suffering 4–12 hour lead response delays, with zero ability to integrate their Zoho CRM data or brand knowledge into conversations. Customers waited hours for replies, human agents were overwhelmed, and every BSP call incurred additional markup on Meta's standard message rates — with no path to automation.

The Solution

Our Approach

Built a custom enterprise WhatsApp Business inbox directly on Meta's Cloud API, replacing the BSP entirely. Every inbound Meta webhook is immediately enqueued via BullMQ and acknowledged with 200 OK, then processed asynchronously — meeting Meta's strict 2-second SLA while decoupling ingestion from AI processing. A ProviderFactory with a shared IAIProvider interface allows Claude, GPT-4o, and Gemini to be swapped at runtime. Diagnosed and fixed a live hallucination bug by redesigning the pipeline to use native function-calling — the agent now issues typed tool calls that fetch real Zoho data before composing any reply. Self-hosted Qdrant replaced external Pinecone, eliminating external billing. TOTP 2FA, Redis caching for Zoho rate limits, and a FIFO CI deployment lock were all built from scratch.

Details

Project Information

Client

MaxFate Private Limited

Timeline

May 2026 – Jun 2026

Role

Sole Developer & Technical Lead

Technologies

25 Technologies

Tech Stack

25 Technologies

TypeScript

Node.js

Express.js

Bun

Next.js

React

Socket.io

BullMQ

Redis

MongoDB

Mongoose

Qdrant

OpenAI API

Anthropic Claude

Google Gemini

Meta WhatsApp Cloud API

Zoho CRM

Zoho Books

AWS EC2

AWS ECR

Cloudflare R2

Docker

GitHub Actions

Turborepo

Zod

Outcomes

Results & Impact

#01

40–70% monthly cost reduction — eliminated WATI/AISensy BSP subscriptions; now paying only Meta's standard conversation rate

#02

96%+ reduction in lead response time — from 4–12 hours to <30 seconds via AI auto-response, enabling 24/7 support

#03

89.7% AI self-service rate in first 16 days — only ~10% of threads needed human intervention

#04

3,652 messages processed across 273 conversations with 99.8% delivery reliability in 16 days live

#05

24.6 million LLM tokens processed with zero downtime

#06

Built and shipped in 3 weeks from zero — 163 commits, single developer

Project Gallery

WhatsApp Inbox — AI Auto-Response with Zoho CRM Sidebar

Facebook Messenger Inbox — Auto AI Mode

Instagram DM Inbox — Auto AI Mode

AI Configuration — Brand Prompt & RAG Similarity Sliders

AI Configuration — Multi-Provider Routing (OpenAI/Claude/Gemini)

AI Configuration — Qdrant Vector Knowledge Base Ingestion

AI Configuration — Real-Time Context Simulation Playground

BullMQ Queue Infrastructure Monitor (4 Queues)

Meta WhatsApp Template Broadcast Console

Create WhatsApp Template — Multi-Step Modal

Security Settings — Operator Profile & TOTP 2FA Setup

Token Cost Analytics — 30-Day Spend by Provider/Model

More Projects

OTHERS

ObservabilityOS — AI-Native DevOps Intelligence & Observability Platform

AI-native log analytics and incident response platform with real-time streaming, statistical anomaly detection, and multi-provider LLM diagnostics.

OPEN SOURCE

Sentinel — Adaptive Distributed Load Control Microservice

Open-source adaptive HTTP reverse proxy in TypeScript/Bun, sustaining ~6,800 req/s at p99 41ms baseline. Protects distributed backends from traffic spikes and cascading failures via hybrid rate limiting, distributed circuit breaking, and real-time pressure-based throttling.