Tanush Sharanarthi
Staff Software Engineer @ IBM
Santa Clara, California, United States
Actions
Tanush Sharanarthi is a Staff Software Engineer at IBM Silicon Valley Labs working at the intersection of artificial intelligence, software engineering, and intelligent systems. His work focuses on agentic AI, large language models, multi-agent orchestration, AI-powered developer tooling, and building reliable production-grade automation systems.
He has designed and built AI-driven architectures that combine specialized agents, tool-based workflows, orchestration layers, and backend systems to solve complex engineering problems. In parallel with his industry work, Tanush is an active researcher and author, publishing work in areas including AI agents, intelligent software systems, and AI-powered software engineering. He also serves as a technical reviewer for research papers and books and has judged multiple hackathons, contributing to the broader developer and research community.
A former Toastmaster and active speaker, Tanush enjoys translating advanced technical ideas into practical, accessible concepts for engineering audiences. His current interests include production AI reliability, agent infrastructure, LLM systems, autonomous software agents, and the future of AI-native software engineering.
Area of Expertise
Topics
Federated Learning Meets Enterprise AI: Scaling Intelligence Without Sharing Data
In an era where data privacy and distributed intelligence are critical, Federated Learning (FL) offers a transformative way for enterprises to train AI models collaboratively without moving or exposing sensitive data. This talk explores how organizations can leverage Asynchronous Hierarchical Federated Learning (FedAH) to overcome the scalability and communication bottlenecks of traditional FL systems. Drawing from real-world enterprise scenarios, it demonstrates how FL can enable secure, low-latency collaboration across diverse environments whether on-premise, edge, or cloud while maintaining compliance with strict governance policies. Attendees will gain insights into designing efficient FL pipelines, handling heterogeneous data sources, optimizing communication rounds, and applying these techniques to large-scale AI workloads in industries such as finance, healthcare, and manufacturing. This session bridges the gap between research and practical deployment, showing how enterprises can scale intelligence without sharing data.
AI Agents in Software Engineering: Can Autonomous AI Teams Build and Maintain Code
AI-powered coding assistants like GitHub Copilot have changed how developers write code, but what if AI could go beyond assisting and work together as a team? This talk explores the potential of multi-agent AI systems, where different AI agents take on specialized roles—one writing code, another reviewing it, a third optimizing performance, and another refactoring—to collaboratively build and maintain software with minimal human input. We’ll dive into real-world applications, emerging research, and the challenges of AI-driven development, from debugging AI-generated code to ensuring reliability. Attendees will gain insight into the future of AI-powered software engineering, whether AI can function as independent development teams, and what this means for the role of human engineers.
git bisect for AI Agents: Replay and Execution Diffing for Production Failures
When traditional software breaks, we have logs, traces, debuggers, and tools like git bisect. When an AI agent breaks, reproducing the failure can be much harder: the model may choose a different tool, take a different routing path, receive changed external data, or hand the task to a different agent.
This session presents a practical architecture for record-and-replay debugging of production AI agents. Every model invocation, context mutation, routing decision, MCP tool call, tool response, agent handoff, and side effect is captured as an execution graph rather than a flat log.
Using these traces, we will explore two debugging techniques. Execution diffing identifies the first meaningful divergence between successful and failed runs. Counterfactual replay keeps the original request and environment fixed while swapping one component, such as the model, prompt, routing policy, or tool, to determine whether a proposed change would actually have prevented the failure.
The session will walk through a multi-agent workflow that fails because of an incorrect routing decision, replay the original execution without repeating external side effects, test an alternative orchestration policy in shadow mode, and identify the decision responsible for the regression.
We will also cover the distributed-systems techniques that make this possible, including event sourcing, idempotency, trace propagation, immutable execution logs, side-effect isolation, and MCP-level instrumentation.
Attendees will leave with a concrete blueprint for answering four questions every production agent platform eventually faces: What failed? Where did it fail? Can I reproduce it? Would my fix actually work?
IBM Silicon Valley Labs - Cloud Education Seminar
Webinar about Basics of Cloud Computing to new hires
CAIT Conference
Presented research on hierarchical federated-learning architectures designed to improve communication efficiency and scalability in distributed ML systems.
BDIIE
Presented research on applying AI-driven digital twins to infrastructure monitoring, predictive analysis, and resilience.
MLCIPR
Presented research in machine learning and intelligent systems as part of the conference’s technical program.
Johns Hopkins Carey Business School - Webinar Speaker
Discussed AI and software-engineering trends, professional experience, career development, and the transition from graduate education to industry.
World SUAS Conference
Presented an AI-agent-based recommendation framework designed to adapt recommendations using user preferences and contextual signals.
ICAIDE
Presented a self-learning multi-agent architecture for real-time software analysis combining retrieval augmentation, reinforcement learning, and specialized AI agents.
AI DevSummit 2025 Sessionize Event
Johns Hopkins University - Alumni Speech
Spoke with MSIS/data-science students about software engineering, AI, industry experience, career development, and applying graduate-school skills in technology roles.
IBM Data & Technology Summit
Presented on AI-assisted natural-language-to-SQL capabilities for Db2, including the use of language models to generate database-optimized SQL and considerations around fine-tuning, security, and enterprise deployment.
EducationUSA Hyderabad / Y-Axis Foundation - International Education Week Speaker
Spoke about the Johns Hopkins Carey MSIS experience, graduate education in the United States, and transitioning from university into the technology industry.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top