Mohit AI - Inbound SDR Platform / www.mohit-ai.com
Built by a Founding Engineer/PM who understands AI SDR workflows
📖 Read Case Study • 🏗️ System Architecture • 🤖 AI Strategy • 📊 Product Metrics
This isn't just a code repository - it's a complete demonstration of how I approach building AI products from first principles: market research → product strategy → technical architecture → production implementation.
Perfect for: Early-stage AI SDR startups looking for founding engineers who can wear multiple hats (PM + AI Engineer + Backend Engineer).
73% of inbound B2B leads never get contacted.
- Average response time: 42 hours (industry standard)
- By then, 78% have engaged with competitors
- Weekend leads have only 8% contact rate
- Result: $150-$300 in marketing spend wasted per lead
Why? Traditional SDR teams can't scale to meet modern buyer expectations for instant response.
An intelligent platform that responds to every inbound lead in <5 minutes with:
- Natural voice calls using ElevenLabs synthesis
- Real-time BANT qualification (Budget, Authority, Need, Timeline)
- 68% call connection rate (vs 50% industry avg)
- OpenAI GPT-4 (primary) + Google Gemini (fallback)
- 99.95% uptime through automatic provider switching
- 87% qualification accuracy vs human SDR review
- Live transcription as the AI speaks
- Instant sentiment analysis and insight generation
- WebSocket-powered dashboards (45ms P95 latency)
- Bi-directional sync with HubSpot & Salesforce
- Automatic lead enrichment and routing
- Full conversation history in CRM
This project showcases expertise across three disciplines:
- Full PRD with user stories, RICE prioritization, GTM strategy
- Metrics Framework - North Star metric, KPI dashboards, A/B testing
- User persona development based on industry research
- Product-market fit hypothesis validation
- Multi-provider AI strategy with automatic fallback
- Prompt engineering & versioning (89% → 92% BANT accuracy through iteration)
- Cost optimization: $0.68 → $0.42 per call (-38%) through caching & model selection
- AI evaluation framework (accuracy, latency, cost monitoring)
- Scalable system architecture - 1,000 req/sec, 5,000 WebSocket connections
- Real-time communication (WebSocket vs polling - see ADR)
- Database design (PostgreSQL + Prisma) with query optimization
- Security (JWT auth, RBAC, rate limiting, encryption)
Architecture Decision Records document every major technical choice:
-
- Why: Single provider = single point of failure
- Solution: OpenAI primary, Google Gemini fallback
- Impact: 99.95% uptime, $500/month cost savings
-
- Why: Real-time transcription needs sub-second latency
- Solution: Socket.io for bidirectional communication
- Impact: 45ms latency vs 2s polling, 90% bandwidth savings
-
- Why: Balance dev speed vs performance
- Solution: Prisma for type safety + migration management
- Impact: 3x faster schema iteration, 7ms acceptable latency trade-off
📊 View Interactive Diagrams - 8 detailed Mermaid diagrams including system architecture, AI call flow, multi-provider fallback, WebSocket architecture, database schema, and more.
Frontend (React)
↓ HTTPS/WSS
Express API + Socket.io Server
↓
┌─────────────────────────────────────────────┐
│ AI Service Factory (Multi-Provider) │
│ ├─ OpenAI GPT-4 (Primary) │
│ ├─ Google Gemini (Fallback) │
│ └─ Circuit Breaker + Health Monitoring │
└─────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────┐
│ Data Layer │
│ ├─ PostgreSQL (Prisma ORM) │
│ ├─ Redis (Cache + Queue) │
│ └─ S3 (Call Recordings, Transcripts) │
└─────────────────────────────────────────────┘
↓
External APIs: Twilio, ElevenLabs, HubSpot
Performance Benchmarks:
- API Throughput: 1,000 req/sec (single instance)
- WebSocket Connections: 5,000 concurrent
- Database Queries: P95 latency 85ms (with indexing)
- AI Call Cost: $0.42/call (optimized from $0.68)
Mohit-AI-Backend/
├── docs/ # Comprehensive documentation
│ ├── product/ # PRD, metrics, roadmap
│ ├── architecture/ # System design, API specs
│ ├── ai-engineering/ # AI strategy, prompts
│ └── decisions/ # ADRs for key tech choices
├── src/
│ ├── core/ # Business logic
│ │ ├── domain/ # Entities (Lead, AICall)
│ │ └── usecases/ # App layer (QualifyLead)
│ ├── ai/ # AI-specific modules
│ │ ├── providers/ # OpenAI, Google, factory
│ │ ├── prompts/ # Versioned prompts
│ │ └── evaluation/ # Quality monitoring
│ ├── api/ # HTTP + WebSocket
│ │ ├── routes/
│ │ ├── controllers/
│ │ └── middleware/
│ └── infrastructure/ # External integrations
│ ├── database/ # Prisma
│ ├── cache/ # Redis
│ └── queue/ # Bull
├── prisma/ # Database schema & migrations
├── CASE_STUDY.md # Full project narrative
└── README.md # You are here
Code Highlights:
- Clean separation of concerns (DDD principles)
- Type-safe database queries (Prisma)
- Comprehensive error handling & logging
- Production-ready security (JWT, rate limiting, encryption)
Designed comprehensive analytics for AI SDR platform:
Lead → Opportunity Conversion Rate
- Target: 15% (vs industry avg 8%)
- Speed: P90 response time <5 minutes
- Quality: 87% AI qualification accuracy
- Scale: 100+ concurrent AI calls
- Cost: $80 cost per qualified lead (vs $280 industry avg)
- Executive: Conversion funnel, ROI, pipeline velocity
- SDR Manager: Lead queue, AI performance, rep productivity
- AI Ops: Provider health, cost tracking, quality scores
Node.js 18+, PostgreSQL 15+, Redis 7+
API keys: OpenAI, ElevenLabs, Twilio (optional for full demo)# Clone repository
git clone https://github.com/Mohit4022-cloud/Mohit-AI-Backend.git
cd Mohit-AI-Backend
# Install dependencies
npm install
# Set up environment variables
cp .env.example .env
# Edit .env with your API keys
# Run database migrations
npx prisma migrate deploy
npx prisma generate
# Start development server
npm run devServer runs at http://localhost:5000
POST /api/leads # Create new lead
POST /api/ai-calls/initiate # Initiate AI call
GET /api/ai-calls/:id # Get call details
GET /api/analytics/dashboard # Metrics dashboard
WS /socket.io # WebSocket for real-time updates
- 📋 Product Requirements Document - Full product spec, user stories, GTM strategy
- 📊 Metrics Framework - North Star metric, KPIs, dashboard designs
- 📖 Case Study - Complete project narrative
- 🤖 AI Strategy & Architecture - Provider selection, prompt engineering, cost optimization
- 🔧 Multi-Provider AI Decision - Fallback architecture rationale
- 🏗️ System Design - Architecture, data flow, scalability analysis
- ⚡ WebSocket Architecture - Real-time communication design
- 💾 Database ORM Choice - Prisma vs alternatives
✅ Multi-provider AI saved the project during OpenAI outage (Aug 2025) ✅ WebSockets enabled sub-second real-time updates (vs 2s polling lag) ✅ Prisma accelerated development by 3x (worth the marginal perf cost)
See: Full Case Study for detailed learnings
This project proves I can:
Think Like a Founder:
- Research market → identify pain point → design solution → ship product
- Make pragmatic trade-offs (cost vs quality, speed vs perfection)
Build Production-Ready AI:
- Multi-provider strategy (not just POC with one API)
- Cost optimization through caching & model selection
- Quality monitoring & continuous improvement
Architect for Scale:
- WebSocket architecture supporting 5,000+ connections
- Horizontal scaling plan (1 → 3 → 10+ instances)
- Database optimization (indexes, connection pooling)
Execute Quickly:
- Built full stack (product + backend + AI) in 3 months solo
- Comprehensive documentation demonstrates thinking, not just code
Want to discuss this project or AI SDR opportunities?
- 📧 Email: mohit@mohit-ai.com
- 💼 LinkedIn: linkedin.com/in/mohittiwari
- 🐙 GitHub: github.com/Mohit4022-cloud
Next Steps:
- 📖 Read the Case Study for full project narrative
- 🏗️ Explore the Architecture Docs to see system design thinking
- 🤖 Check out AI Strategy for prompt engineering & cost optimization
This project is proprietary software created for portfolio demonstration purposes.
Industry Data Sources:
- InsideSales.com - Lead Response Study (2024)
- Bridge Group - SDR Metrics Report (2025)
- LeanData - Pipeline Generation Benchmark (2024)
Tech Stack:
- OpenAI for GPT-4 API
- Google for Gemini API
- ElevenLabs for voice synthesis
- Twilio for communication infrastructure
Built with product thinking, AI expertise, and backend rigor
Last Updated: October 17, 2025