Amith Reddy Ravuru
Staff Platform Architect
Actions
I am an engineering leader at Reltio, an SAP company, with more than 15 years of experience building distributed systems, cloud-native platforms, and operational tooling for enterprise environments. My work focuses on enterprise data platforms, multi-cloud SaaS operations, observability, reliability, and trustworthy agentic AI systems.
My platform work has included self-service tenant lifecycle automation across AWS, Azure, and GCP, tenant-centric operational visibility for faster troubleshooting, and grounded AI-assisted workflows for onboarding and incident operations. My recent interests center on production-grade AI systems: how agents behave under load, how tool use should be validated, how AI workflows should be observed, and how governance and human review can be built into autonomous data systems.
Across roles at Reltio, GoTo, Citrix, and Huawei, I have worked on reliability-sensitive enterprise and telecom systems. I also wrote about modern data engineering and governed AI, including his Data City series and my article on Agentic Data Unification. As a speaker, I bring a practitioner perspective that combines architectural depth, operational realism, data governance, and evidence-based evaluation of technical impact.
The Token Economics of Enterprise Agents: A Modeled Path to $300,000 in Annual Savings
Agentic AI costs can grow unexpectedly when workflows repeatedly send large prompts, retrieve excessive context, invoke expensive models, transfer verbose tool results, or allow agents to continue reasoning after a useful answer is available. These inefficiencies are difficult to detect when teams measure only total API spending rather than the economics of each workflow step.
This presentation develops a transparent, hypothetical enterprise scenario involving 10 million agent workflows per year. At illustrative blended rates of $2 per million input tokens and $8 per million output tokens, a baseline workload averaging 20,000 input and 2,000 output tokens per workflow would cost approximately $560,000 annually. Reducing average input usage to 7,000 tokens and output usage to 1,500 tokens would lower the modeled cost to approximately $260,000, representing a potential annual saving of $300,000 (~54%).
The session explains how an organization could achieve these reductions through semantic context selection, retrieval pruning and reranking, prompt compression, conversation summarization, tool-result compaction, model routing, caching, bounded agent loops, early termination, and token budgets. It also shows how quality evaluations and business-outcome metrics can prevent cost optimization from degrading accuracy, safety, or user experience.
Attendees will leave with a model-independent framework for measuring cost per successful outcome, locating token waste, prioritizing optimizations, and building financial guardrails for enterprise agent platforms.
Learning from Every Trace: Using OpenTelemetry to Improve Agentic Workflows
Agentic systems can fail, slow down, or overspend in ways that are difficult to diagnose from the final response alone. A single request may involve planning, retrieval, model calls, tool execution, agent handoffs, retries, policy decisions, and evaluation steps. Without end-to-end telemetry, teams cannot reliably determine where quality was lost, latency accumulated, or tokens were wasted.
This session demonstrates how OpenTelemetry can provide a common observability model across the agent lifecycle. It shows how traces and spans can represent prompts, model selections, retrieved context, tool calls, agent transitions, retries, token consumption, estimated cost, latency, evaluation results, and final business outcomes. These signals can expose unnecessary reasoning loops, oversized context, redundant tool calls, ineffective retrieval, expensive model choices, and recurring failure paths.
The session then connects production telemetry to the development lifecycle. Trace-derived findings can generate optimization proposals, create regression tests, inform prompt and workflow revisions, and pass through evaluation, artifact, approval, canary, and rollback gates before release. This creates a governed improvement loop rather than allowing agents to modify production workflows without oversight.
Attendees will learn how to use telemetry to improve reliability, adapt routing and context strategies, reduce autonomous-agent costs, and establish measurable evidence that each workflow revision is safer, faster, or more effective than the version it replaces.
Learning from Every Trace: Using OpenTelemetry to Improve Agentic Workflows
Agentic systems can fail, slow down, or overspend in ways that are difficult to diagnose from the final response alone. A single request may involve planning, retrieval, model calls, tool execution, agent handoffs, retries, policy decisions, and evaluation steps. Without end-to-end telemetry, teams cannot reliably determine where quality was lost, latency accumulated, or tokens were wasted.
This session demonstrates how OpenTelemetry can provide a common observability model across the agent lifecycle. It shows how traces and spans can represent prompts, model selections, retrieved context, tool calls, agent transitions, retries, token consumption, estimated cost, latency, evaluation results, and final business outcomes. These signals can expose unnecessary reasoning loops, oversized context, redundant tool calls, ineffective retrieval, expensive model choices, and recurring failure paths.
The session then connects production telemetry to the development lifecycle. Trace-derived findings can generate optimization proposals, create regression tests, inform prompt and workflow revisions, and pass through evaluation, artifact, approval, canary, and rollback gates before release. This creates a governed improvement loop rather than allowing agents to modify production workflows without oversight.
Attendees will learn how to use telemetry to improve reliability, adapt routing and context strategies, reduce autonomous-agent costs, and establish measurable evidence that each workflow revision is safer, faster, or more effective than the version it replaces.
Learning from Every Trace: Using OpenTelemetry to Improve Agentic Workflows
Agentic systems can fail, slow down, or overspend in ways that are difficult to diagnose from the final response alone. A single request may involve planning, retrieval, model calls, tool execution, agent handoffs, retries, policy decisions, and evaluation steps. Without end-to-end telemetry, teams cannot reliably determine where quality was lost, latency accumulated, or tokens were wasted.
This session demonstrates how OpenTelemetry can provide a common observability model across the agent lifecycle. It shows how traces and spans can represent prompts, model selections, retrieved context, tool calls, agent transitions, retries, token consumption, estimated cost, latency, evaluation results, and final business outcomes. These signals can expose unnecessary reasoning loops, oversized context, redundant tool calls, ineffective retrieval, expensive model choices, and recurring failure paths.
The session then connects production telemetry to the development lifecycle. Trace-derived findings can generate optimization proposals, create regression tests, inform prompt and workflow revisions, and pass through evaluation, artifact, approval, canary, and rollback gates before release. This creates a governed improvement loop rather than allowing agents to modify production workflows without oversight.
Attendees will learn how to use telemetry to improve reliability, adapt routing and context strategies, reduce autonomous-agent costs, and establish measurable evidence that each workflow revision is safer, faster, or more effective than the version it replaces.
Making Business Context Machine-Readable: Semantic Ontologies for AI-Ready Data
Most organizations have accumulated fragmented source schemas, legacy data models, inconsistent terminology, and disconnected definitions of important business entities. These inconsistencies become especially consequential when generative and agentic AI systems use enterprise data to answer questions, make recommendations, or take action.
This presentation explains how semantic ontologies can transform fragmented data structures into connected, machine-readable business context. It follows the architectural journey from source schemas and legacy models to canonical entities, attributes, relationships, business definitions, and knowledge graph representations. It also clarifies the differences among schemas, taxonomies, business glossaries, semantic models, knowledge graphs, and vector retrieval. It shows how these capabilities complement one another rather than compete.
Using a practical, vendor-neutral example, the session will demonstrate how ontology mapping, validation, versioning, stewardship, provenance, and policy-aware context can improve AI readiness. Attendees will learn how to establish shared business meaning that can be consistently understood and applied by data practitioners, applications, and AI agents.
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