Shah Khurram

Designed and implemented enterprise multi-agent systems with autonomous task execution, goal-driven behaviors, and role-based coordination across distributed workflows. Architected dual-layer vector memory combining agent-specific memory in Pinecone with a global cross-agent knowledge base, plus an intelligent retrieval strategy that prioritizes local memory and falls back to global knowledge when needed. Built tool-calling agents with API orchestration, enabling intelligent function execution, external system integration, and reasoning over real-time data. Engineered a provider-agnostic AI orchestration layer with encrypted API key management, supporting multiple LLM providers including OpenAI, Anthropic, Gemini, Groq, Mistral, Meta LLaMA, and Cohere. Designed real-time voice AI assistants delivering human-like conversational responses with low-latency streaming STT/TTS. Built RAG-grounded voice agents tied to agent-specific and global knowledge bases. Implemented per-agent voice configuration with custom voice names, personalities, models, and independent behavioral tuning. Owned end-to-end AWS infrastructure (EC2, RDS, S3) including security, monitoring, autoscaling, and cost optimization. Built and maintained CI/CD pipelines using GitHub Actions. Designed system architectures and database schemas for scalable, high-traffic enterprise applications. Integrated third-party services including Tavily, Cohere, Stripe, Postmark, and Google OAuth 2.0. Led and mentored developers, conducted code reviews, managed delivery timelines, and represented engineering in weekly CEO-level strategy meetings.

Promoted from Trainee Developer (Jan 2023 - Mar 2023) to full-time engineer; balanced full-time work with completing B.Tech under university No-Objection certification. Built an Intercom-style Agentic AI platform featuring multi-agent support, agent-specific vector knowledge bases, embeddable JavaScript widgets, and chat plus voice assistant popups for SaaS-style deployment. Developed scalable backend systems in PHP (CodeIgniter, Laravel) and Node.js for appointment scheduling, workflow automation, and real-time WebSocket communication. Designed privacy-first document intelligence pipelines supporting PDF, DOCX, and OCR-based extraction with structured data validation, auto-filling Hospital Management Systems via on-premises Hugging Face LLMs. Implemented recursive web scraping pipelines with depth-controlled crawling and automated vector database ingestion to power semantic search and RAG retrieval. Integrated third-party APIs including payments, identity verification, email services, and Clio for client-facing workflow automation.

Worked as a trainee developer while completing B.Tech. Assisted in building backend systems and AI pipelines, and contributed to document intelligence and workflow automation projects.