I'm an Enterprise AI Solutions Architect with 14+ years of experience building production-ready agentic AI systems, AI-powered software delivery accelerators, real-time Voice AI, and RAG applications — from architecture and proof-of-concept through enterprise-scale production deployment.
My recent work spans multi-agent orchestration (MCP, LangGraph, coordinator-led delivery frameworks), AI-driven automation (agentic SDLC, testing accelerators, evaluation harnesses), real-time Voice AI (composable STT–LLM–TTS pipelines, multimodal conversational agents), and retrieval-augmented generation — deployed across AWS, Azure, and NVIDIA AI platforms.
👯 Looking to collaborate on agentic AI systems, AI-driven automation, Voice AI architecture, and RAG pipelines 💬 Ask me about multi-agent orchestration, agent evaluation frameworks, Voice AI, retrieval-augmented generation, or NVIDIA/AWS/Azure AI platforms
- Designing enterprise-grade agentic AI systems and multi-agent orchestration workflows
- Building AI-powered automation — agentic SDLC frameworks, testing accelerators, evaluation harnesses
- Building real-time Voice AI agents — composable STT–LLM–TTS pipelines with multimodal, embodied response
- Building RAG pipelines over structured and unstructured enterprise data
- Building secure, MCP-connected agents for enterprise data, dev workflows, and testing
- Exploring the frontier across NVIDIA, AWS, and Azure AI platforms
A coordinator-led multi-agent software delivery framework (Research, Dev, Code Review, Testing, CI/CD agents) integrated via MCP, with Jira/GitLab connectors for git-native, PR-linked workflows. Includes an AI-assisted testing agent (LangGraph + Playwright) that cut manual testing effort by ~40% and scaled to a 50+ entity accessibility-compliance pipeline.
An evaluation harness benchmarking multi-agent and LLM systems on accuracy, tool-use correctness, reasoning quality, latency, and regression — with ground-truth feedback loops driving evidence-based agent design decisions.
A Jira-connected agent built on NVIDIA's agentic framework, with scoped tool access, sandboxed execution, and governance controls. Contributed an upstream open-source PR and published an article on secure multi-agent execution.
A SQL-connected agent exposed via MCP, with LiteLLM as the model-routing/gateway layer and a custom-built session, authentication, and authorization layer. Gives business users on-demand extraction and categorization — summary, sentiment, classification, trend detection — over large structured datasets, no SQL required.
A real-time, composable Voice AI system integrating LiveKit and Pipecat for audio orchestration, NVIDIA Riva/NeMo for ASR/TTS, and BitHuman/Unreal Engine MetaHuman for embodied multimodal response — with WebRTC streaming, session management, and interruption/turn-taking handling. Demonstrated live at NVIDIA GTC 2026.
A real-time conversational Voice AI agent built on NVIDIA PersonaPlex and Nemotron, handling natural-language booking workflows end-to-end — from intent capture through transaction completion.
A production RAG pipeline over enterprise engineering drawings and specifications, using OCR-based extraction, Google ADK, and LangChain — with embeddings, reranking, and guardrails over unstructured technical documents, and LangSmith tracing for observability.
Agent & LLM Stack: LangChain, LangGraph, MCP (Model Context Protocol), A2A, CrewAI, AutoGen, Claude Agent SDK, OpenAI Agents SDK, Gemini ADK, AWS Strands, vector databases, tool integration
Evaluation & Observability: Agent evaluation harnesses, LLM benchmarking, latency/cost evaluation, LangSmith, Langfuse, Arize Phoenix
Voice AI & Real-Time Systems: LiveKit, Pipecat, NVIDIA Riva (ASR/TTS), BitHuman, Unreal Engine MetaHuman, WebRTC, streaming audio, interruption handling, latency optimization
LLMs & Model Development: Foundation models, advanced & multimodal RAG, fine-tuning (LoRA/PEFT), embeddings, reranking, prompt engineering, guardrails, Hugging Face, vLLM, Ollama
Cloud & Platform: AWS Bedrock, AWS AgentCore, Amazon EKS, Azure AI Foundry, Azure ML, NVIDIA AI Enterprise, Kubernetes, Docker, GitOps (Argo CD), CI/CD (GitLab), Linux
NVIDIA AI Stack: NeMo, Nemotron, NeMo Agent Toolkit, NIM, TensorRT-LLM, Triton Inference Server, CUDA, GPU inference
Languages: Python, TypeScript/JavaScript, Go, Java, SQL
- NVIDIA Certified Professional: Agentic AI (2026)
- Anthropic Claude Certified Architect (2026)
- NVIDIA Certified Associate: Gen AI LLM (2024)
- NVIDIA AI Advisor (2025)
- Post Graduate Program in AI/ML — University of Texas at Austin
Multi-agent orchestration & MCP infrastructure
AI-driven automation (agentic SDLC, testing)
Real-time Voice AI architecture
Retrieval-augmented generation (RAG)
Cross-platform AI deployment (AWS / Azure / NVIDIA)
Secure, governed enterprise agentic systems
📫 Reach me: LinkedIn | shiva.perumalsamy@gmail.com