Tom Vellavoor Saji

About

Tom Vellavoor Saji

Senior AI Engineer — Azure AI, Agentic Systems, and Enterprise LLM Platforms

I've spent the last four years designing and shipping AI systems that run in production for some of the largest organisations in Qatar — spanning media, education, government, and investment. Before that I studied Electrical & Electronics Engineering, which is where I first got interested in the intersection of systems thinking and software.

I work best at the boundary between research and production: taking ideas from papers and prototypes and turning them into systems that are reliable, observable, and maintainable at scale. Right now I'm especially focused on agentic architectures — multi-agent orchestration, evaluation pipelines, and the infrastructure that makes agents trustworthy in real deployments.

Currently building AI infrastructure at scale across Qatar — spanning enterprise chatbots, real-time avatar platforms, multilingual media AI, and government-grade agentic systems.

Expertise

Platforms, systems, and tools

AI Platforms

  • Azure OpenAI
  • Azure AI Foundry / Foundry Agents
  • Azure AI Search
  • Semantic Kernel
  • LangChain
  • LangGraph
  • Microsoft Foundry Agent Service
  • Model Context Protocol (MCP)

Cloud & DevOps

  • Azure Container Apps
  • Azure Functions
  • Azure Storage
  • PostgreSQL
  • Cosmos DB
  • Service Bus
  • Entra ID & RBAC
  • Docker
  • Azure DevOps
  • GitHub Actions

Application Engineering

  • Python
  • FastAPI
  • TypeScript
  • React / Next.js
  • REST APIs
  • Microservices
  • WebSockets
  • Real-time AI systems

Selected work

Production systems

Botaina — Enterprise AI Chatbot (Qatar Foundation)

  • Enterprise chatbot built with Azure OpenAI and Azure AI Search
  • Handles 3,000+ daily queries
  • Achieves ~90% resolution rate
  • Built secure private architecture using Entra ID and RBAC
  • Supports multi-language RAG pipelines

Architecture

Users → API Layer → RAG Orchestrator → Azure AI Search + GPT-4o → Guardrails + RBAC

Al Jazeera — Multilingual Video AI Platform

  • Hybrid Whisper + Azure transcription pipeline
  • Supports 10+ languages for transcription and translation
  • Includes AI content moderation in the processing loop
  • Reduced editorial review time from hours to minutes

Architecture

Ingest → Speech-to-Text (Whisper + Azure) → Translation → Moderation → Editorial Delivery

Real-Time AI Avatar Platform

  • WebRTC video streaming with low-latency response loops
  • WebSocket audio pipelines for conversational flow
  • Semantic Kernel tool calling integrated with business systems
  • Azure Speech TTS avatars for production deployments
  • Supports 50+ concurrent sessions across multiple products

Architecture

Client ↔ WebRTC Session ↔ Real-Time Agent Core ↔ Tooling + TTS Avatar + Enterprise APIs

Government AI Agents (Foundry)

  • Private agentic AI systems for government entities
  • Built on Microsoft Foundry Agents with secure tenancy boundaries
  • Bing Custom Search grounding and multi-agent orchestration
  • Secure Entra ID access controls for enterprise-grade governance
  • Used by 200+ users across multiple departments

Architecture

Entra ID Auth → Foundry Agent Mesh → Grounding Services → Department Workflows

Side projects

Independent builds

  • AI Agent Builder Platform
  • LangChain RAG Agent

Credentials

Certifications & education

  • Microsoft Certified: Azure AI Engineer Associate
  • Microsoft Certified: Azure Developer Associate
  • B.E Electrical & Electronics Engineering — Anna University (2020)