Shah Khurram

Designed complete system architectures and database schemas for scalable, high-traffic enterprise applications. Built and maintained CI/CD pipelines using GitHub Actions, enabling automated builds, deployments, rollbacks, and environment isolation. Owned and managed AWS infrastructure (EC2, RDS, S3), including security, monitoring, scaling, and cost optimization. Designed and implemented enterprise-level multi-agent systems with autonomous task execution, goal-driven behaviors, and role-based coordination. Built tool-calling agents with API orchestration, enabling intelligent function execution and external system integration. Architected dual-layer memory systems: agent-specific vector memory (Pinecone) for isolated knowledge and global vector knowledge base for cross-agent intelligence. Implemented intelligent retrieval strategies prioritizing agent-level memory with fallback to global knowledge when required. Built provider-agnostic AI orchestration layer with encrypted API key management, supporting OpenAI, Anthropic, Gemini, DeepSeek, Groq, Mistral, Meta (LLaMA), Cohere, Perplexity, Together AI, Qwen, AI21, and Grok. Designed and implemented real-time voice AI assistant agents with human-like conversational responses comparable to ChatGPT. Built RAG-powered voice agents grounded in agent-specific and global knowledge bases with low-latency streaming responses. Implemented per-agent voice configuration allowing custom voice names, personalities, models, and independent behavioral tuning. Integrated Tavily Web Search API for real-time browsing and reasoning; Cohere embeddings for cost-efficient vectorization; Stripe for subscriptions and usage-based billing; Postmark for transactional emails; Google OAuth 2.0 for authentication. Led and mentored developers, conducted code reviews, managed delivery timelines, and participated in weekly CEO-level strategy meetings.

Developed scalable backend systems using PHP (CodeIgniter, Laravel) and Node.js for appointment scheduling, workflow automation, and real-time communication (WebSocket). Built an Agentic AI platform like Intercom featuring multi-agent support, agent-specific vector knowledge bases, embeddable JavaScript widgets, and chat + voice assistant popups. Designed privacy-first document intelligence pipelines supporting PDF, DOCX, OCR-based extraction, structured data validation, and auto-filling Hospital Management Systems using on-premises LLMs (Hugging Face). Implemented recursive web scraping pipelines with depth-controlled crawling and automated vector database ingestion for semantic search and RAG. Integrated third-party APIs including payments, identity verification, email services, and Clio (legal firm management system).