AI Sales Executive Platform (ASEP)

An industry-grade, autonomous pre-sales qualification agent and real-time CRM monitoring platform designed for Indian IT consultancy markets.

This platform uses an event-driven web-sockets architecture, a custom state-machine orchestrator powered by the Gemini Flash Lite SDK, semantic RAG vector querying, and custom transactional mail dispatches to qualify leads, automate proposals, and schedule callbacks dynamically.

Live Demo

The redesigned client is deployed on Vercel: https://client-blue-nu-26.vercel.app

The public demo runs without a backend by default. Set VITE_API_URL to a deployed API URL to enable live Socket.io conversations, lead qualification, and executive handoff.

---

🛠️ Tech Stack & Architecture

graph TD
    Client[React + Vite Frontend Client] <-->|Socket.io Event Channel| Server[Express Node.js Server]
    Client <-->|REST API JSON Gateway| Server
    Server <-->|ODM Layer| Mongo[(MongoDB Atlas Database)]
    Server <-->|Semantic Queries| Qdrant[(Qdrant Vector Database)]
    Server <-->|Structured JSON Prompting| Gemini[Google Gemini AI Engine]
    Server -->|Nodemailer SMTP| SMTP[SMTP Transactional Email Service]

Loading

1. Backend Architecture (Express + TypeScript)

• Design Pattern: State Machine Orchestrator
  • The pre-sales conversations follow a state pattern: GREETING ➔ DISCOVERY ➔ REQUIREMENT_COLLECTION ➔ BRAINSTORMING ➔ QUALIFICATION ➔ PACKAGE_RECOMMENDATION ➔ PROPOSAL_GENERATION ➔ HANDOFF ➔ CLOSED.
  • The orchestrator dynamically moves stages as BANT parameters are qualified.
• Design Pattern: Event Broker Pattern (Sockets)
  • Uses Socket.io to stream real-time updates (typing indicators, chat takeover signals, notification alerts) across rooms.
• Design Pattern: Repository Pattern (Mongoose ODM)
  • Interfaces with Mongoose schemas (Lead, Conversation, Message, Proposal, Meeting) to ensure transactional consistency across documents.

2. Frontend Architecture (React + Vite + Tailwind CSS)

• Single Page Dashboard & Client Widgets:
  • Visitor Chat Widget: Responsive slide-up drawer showing bot status, interactive suggestion pills, typing dots, and downloadable PDF cards.
  • Executive CRM Workspace: A split-screen dashboard displaying a qualified leads pipeline, a live takeover chat console, and a CRM panel with dial/email shortcuts, callback timers, and quick notes.

---

💎 Advanced Real-World Features

1. 🇮🇳 Local Market Competitive Pricing

• Computes real-time project estimates in Indian Rupees (INR - ₹) based on competitive local consultancy scales.
• Standardized pricing rules:
  • Web Designing & Development: Basic Showcase sites range from ₹25,000 to ₹50,000. Professional dynamic apps range from ₹1 Lakh to ₹2.5 Lakhs. Custom SaaS apps range from ₹3 Lakhs to ₹8 Lakhs+.
  • UI/UX Designing: Figma wireframes and branding prototypes range from ₹25,000 to ₹1 Lakh.
  • Mobile Applications: iOS & Android apps (Flutter/React Native) range from ₹3.5 Lakhs to ₹10 Lakhs.
  • SEO: Ranking, backlinks, page-speed check: ₹15,000 to ₹40,000/month.
  • Content Writing: High-quality SEO copywriting: ₹10,000 to ₹25,000/month.
  • Digital Marketing & PPC Ads Shield: Campaign setup & ad fraud fraud-prevention block shield: ₹25,000 to ₹60,000/month.

📅 2. Natural Language Scheduling (NLP Date Parser)

• The agent asks the client for their preferred callback date and time.
• The parser converts natural descriptions (e.g. "tomorrow at 5 PM", "afternoon", "Monday morning") into exact UTC database timestamps.

📧 3. Deferred Transactional Emailing

• If the user shares their email address after a callback has been scheduled, the backend captures the save event and dispatches the tailored project brief (including their timeline, budget, features, and Meet link) directly to their inbox.

💹 4. Adaptive Market Pricing

• Pricing lives as versioned data, not code: a PricingConfig rate card that the AI reads fresh on every conversation, so quotes always use the current ranges.
• A daily background job re-tunes each service's price multiplier within a bounded corridor (±30% by default) based on:
  • Win/loss feedback — proposal outcomes recorded via PATCH /api/v1/proposals/:id/status (win rate > 60% nudges prices up, < 35% nudges them down, minimum 5 samples per service).
  • Competitor position — CompetitorPrice records; prices ease down when we sit well above the market median and may rise when below.
  • Demand trend — qualified-lead volume over the last 7 days vs. the previous 7.
• Every proposal records which pricing version generated it (pricingVersion + matched services) so the feedback loop can attribute outcomes to price points. Stale SENT proposals auto-expire after 30 days.
• Updating over time: the job runs at boot and then every PRICING_REFRESH_HOURS (default 24h). Record competitor observations via POST /api/v1/competitors ({ service, competitor, price }), inspect the live rate card via GET /api/v1/pricing/rates, and trigger an immediate re-tune via POST /api/v1/pricing/recompute. Proposal decisions (PATCH /api/v1/proposals/:id/status) also trigger a recompute on the spot.
• Automatic market capture: a market scanner periodically fetches configured competitor pricing pages (MarketTarget records, managed via /api/v1/market-targets) and extracts INR rates — JSON-LD structured data first, plain ₹ amounts as fallback. Observations land in CompetitorPrice and feed the next re-tune. Run it on demand with POST /api/v1/market-targets/scan; the check interval is MARKET_SCAN_HOURS (default 24h) and each target has its own refresh window (intervalHours, default weekly).

---

🚀 Easy Local Setup

We have included an automated setup utility script to prepare your workspace in seconds.

Prerequisites

• Node.js: v18+
• npm: v9+
• MongoDB: Atlas cloud connection string or local running instance.

Setup Instructions

1. Clone this repository to your local directory.
2. Grant execution permissions and run the setup script:

   chmod +x setup.sh
   ./setup.sh

   This script will verify prerequisites, copy .env.example templates, install dependency trees, and pre-compile the server and client builds.
3. Open server/.env and update your MONGO_URI, GEMINI_API_KEY, and JWT_SECRET (the server refuses to boot with the placeholder secret). For production, also set PUBLIC_BASE_URL to your deployed API URL so proposal PDF links work.
4. Launch the application:
   • Start Backend API: cd server && npm run start
   • Start Client Dev: cd client && npm run dev
5. Open http://localhost:5173 to interact with the platform!

☁️ Deployment (free tier, judge-demo ready)

The recommended free setup keeps the whole product alive reliably: Vercel for the client, Koyeb for the long-running server (WebSockets + scheduled pricing/market jobs need a real process, not serverless), and MongoDB Atlas free tier for data. Qdrant is optional — the RAG layer falls back to built-in chunks when it is offline.

1. Server → Koyeb (koyeb.yaml at repo root)

1. Sign up at koyeb.com (GitHub login works).
2. Import the repo dgexplores/smart-chatbot → it picks up koyeb.yaml (builds server/Dockerfile, runs on port 5001, health-checked at /health).
3. Set service variables (secrets are never committed):
   • JWT_SECRET (required — generate with openssl rand -hex 32)
   • MONGO_URI from your Atlas cluster
   • PUBLIC_BASE_URL = the Koyeb service URL (makes proposal PDF links work)
   • CLIENT_URL = the Vercel URL below (CORS)
   • MOCK_LLM=false + GEMINI_API_KEY for real AI (free key at aistudio.google.com)
4. SEED_DEMO_DATA=true is set in koyeb.yaml — on first boot it seeds 10 realistic leads, conversations, proposals (with win/loss outcomes), meetings, and 8 weeks of competitor market observations so the dashboard looks alive for a showcase.

2. Client → Vercel

1. vercel login, then from client/: vercel --prod (uses client/vercel.json).
2. Set VITE_API_URL to the Koyeb service URL in the project settings and rebuild.

3. Verify

• GET https://<server>/health → { status: 'ok' }
• GET https://<server>/api/v1/pricing/rates → live rate card with multipliers
• Open the Vercel URL → chat widget talks to the real backend via WebSockets.

Alternative hosts

• Railway (scripts/deploy-railway.sh — one-command deploy): note the free plan can hit the resource-provision limit; Hobby ($5/mo) is required for more services.
• Docker anywhere: server/Dockerfile builds the API for any container host.
• Render/Fly: same Dockerfile; expect free-tier sleep/cold-start behaviour.