I build data systems and research how AI agents should act.
My focus: runtime governance and multi-agent orchestration.
In January 2019, I left India at 22. It was my first flight.
I knew nobody where I was going and had not booked even one night in a hotel. As the plane descended, I remember looking down at snow on the houses. Everything felt unfamiliar. I wanted a better life and a chance to build something. I had little idea how to begin.
Techworkify was my first entrepreneurial venture, focused on technology staffing through HyperLedger OÜ in Estonia. COVID and travel restrictions interrupted my plans and prevented me from moving the venture forward as intended.
I later completed my master’s in Entrepreneurship & Innovation at Hult International Business School in San Francisco. In 2025, I founded DataGovernAI LLC.
My background in data engineering also shaped my interest in AI systems. While developing a concept for an AI leasing assistant, I became interested in the decisions behind the conversation.
The concept involved an assistant that would look up availability and help schedule tours. That raised questions about outdated information, incorrect assumptions passed between agents, and actions that required human approval.
Those questions led to my patent application and my focus on runtime governance. I now study how agents coordinate work, check evidence, and involve people in decisions. I pursue these questions through independent research, doctoral studies in information technology, and my HOMININI prototype.
A conceptual view of evidence-bound execution.
An agent identifies the next step.
Is the action permitted?
Is the supporting information sufficient?
Policy Enforcement, Monitoring, and Risk-Based Human Escalation
An action can be permitted and still rely on incomplete or outdated information. My paper examines the evidence checks and human escalation needed at execution time.
Read on ResearchGate ↗The application describes an architecture for coordinating specialized agents with real-time data integration, governance checks, audit records, and fallback mechanisms. It grew out of the virtual leasing use case that first drew me to this problem.
System and Method for Governed Multi-Agent AI Workflow with Real-Time Data Integration and Session-Aware Voice Automation.
Named inventor: Kiran Kumar Yenaganti
Nonprovisional utility application: 19/269,037
Filing date: July 15, 2025
A running interface can still fail to explain what stopped, what happened, or whether any useful work was completed.
HOMININI is the experimental system I use to investigate that gap. This repair addressed a narrow but consequential failure: an owner asked why work had stopped, and the system returned an acknowledgement instead of an answer.
These counters record execution cycles. They do not establish 13 useful outcomes.
The diagnosis path now resolves the current operational goal, reads its recorded state, explains the execution boundary, and preserves uncertainty when the goal is missing or ambiguous. It does not restart work or grant execution authority.
The repair routes the original diagnostic request to diagnosis rather than acknowledgement; handles conflicting history and multiple goals without guessing; denies restricted access before reading host runtime; and avoids an unnecessary historical scan.
The developer-reported results cover the broader gateway and diagnosis helper, not 56 independent demonstrations of productivity. The tests use a temporary local database, controlled fixtures, mocks, and an action callback that records attempted execution. They establish the specified diagnostic behaviors; they do not demonstrate autonomous productivity, general natural-language understanding, comprehensive security, scientific novelty, or superiority over another system.
A recorded live Portal check returned the iteration-limit explanation without starting a mission. The local evidence has not yet been sanitized into a public reproduction package.
My technical background includes data pipelines, transformation, and data quality. These are the foundations that other systems rely on.
In progress · AI emphasis
San Francisco
For research collaboration and technical discussions, reach me on LinkedIn.
Open my LinkedIn profile ↗