AI · Agent Systems · Autonomous
KESEVAN
SEKAR BALAJI
Building systems that act — from drone autonomy to AI agents at enterprise scale.
ABOUT
Software engineer based in Bangalore. Started out building autonomous drones — embedded systems, computer vision, full-stack from firmware to perception. That curiosity about machines that act on their own eventually led to AI agents at enterprise scale.
At Shell, led the first Agent-to-Agent (A2A) implementation on Azure AI Foundry and built multi-agent systems that automate real infrastructure work. Also spent significant time on a problem most teams ignore: how do you give an AI agent a secure, auditable identity?
Before this, research intern at Siemens working on edge AI and generative models for industrial automation.
Currently looking for AI Engineer or SDE roles at product-focused tech companies — particularly in the Azure AI and developer tooling space.
Agentic AI
Azure AI Foundry·A2A protocol·Multi-agent orchestration·LangGraph·Semantic Kernel
Identity & Security
Microsoft Entra ID·Non-human identity·OAuth / OIDC·IAM lifecycle
Backend & Cloud
Python·FastAPI·Azure·Docker·Linux
Autonomous Systems
ROS / Gazebo·OpenCV·UAV firmware·Embedded systems
PROJECTS
Shell projects are architecture-visible on request.
Agent-to-Agent (A2A) Implementation
Shell's first A2A implementation on Azure AI Foundry. Built the protocol layer that allows autonomous agents to communicate, delegate tasks, and coordinate without human-in-the-loop — enabling true multi-agent pipelines at enterprise scale.
Multi-Agent Orchestrator
Orchestration layer routing backend service automation tasks to specialised sub-agents. Reduced manual operational toil across cloud infrastructure by replacing human-driven runbooks with autonomous agent execution.
IAM Onboarding Agent
Agentic solution simplifying application authentication and onboarding against Microsoft Entra ID. Alongside this, researched non-human identity frameworks for secure autonomous agents — a largely unsolved problem in enterprise AI.
Secure Log Analysis Agent
Agent for automated log analysis, anomaly detection, and secure export across cloud infrastructure. Proactive monitoring for failure prediction — reducing mean-time-to-detect on infrastructure incidents.
Industrial CV & AV Simulation Pipeline
Computer vision system for industrial defect detection paired with a statistical generative learning pipeline for synthetic data generation. Used for varied scenario simulation in autonomous vehicle testing at Siemens Technology.
Drone Autonomy Platform
Autonomous multirotor simulation in ROS/Gazebo with computer vision for aerial object detection. Full stack from firmware to perception — built across AEPL's drone division and the Edhitha UAV Team.
EXPERIENCE
MTS Engineer — AI Agent Platform
- Leading Shell's first Agent-to-Agent (A2A) implementation on Azure AI Foundry
- Multi-agent orchestrator for automated backend service actions across cloud infrastructure
- Proactive monitoring agent for anomaly detection and failure prediction
- IAM-focused agent for application authentication and onboarding (Entra ID)
- Research on non-human identity frameworks for secure autonomous agents
Research Intern
- Statistical generative learning pipeline for synthetic training data
- Varied scenario simulation for autonomous vehicle testing
- CV-based industrial defect detection system
- EDA on large operational datasets using LLMs
Research Intern — Advanced Technology
- Led development for the new autonomous drone division
- ROS/Gazebo simulation of autonomous multirotors
- Computer vision for aerial object detection
B.E. Computer Science · GPA 8.89
- Edhitha UAV Team — CV and embedded systems lead
- Computer vision model for aerial object detection
- Dawn Jaeger Tenacity Award
WRITING
ALL POSTS →Implementing A2A at Enterprise Scale: What the Spec Doesn't Tell You
The protocol gaps, identity challenges, and what actually breaks when autonomous agents try to talk to each other at Shell.
Non-Human Identity: The Problem Nobody in Enterprise AI Is Talking About
When an AI agent calls an API, rotates a secret, or delegates a task — whose identity does it use? Why OAuth/OIDC wasn't built for agents.
From ROS to LangGraph: What Drone Autonomy Taught Me About AI Agents
State machines, sensor loops, failure recovery — the parallels between building autonomous UAVs and LLM agent pipelines.
CONTACT
Open to AI Engineer and SDE roles at product-focused tech companies — particularly Azure AI and developer tooling.
Fastest path: LinkedIn DM. Prefer async? Email. Want to see what's being built? GitHub.