Sebastian Esponda

Principal Engineer - Architect Sydney, Australia Engineering since 2000

Principal engineer with experience building production AI agents and evaluation infrastructure. Built an autonomous multi-turn agent testing framework, now the subject of a pending patent. Currently in Atlassian's Central AI group, working on search ranking and relevance for third-party connectors.

Twenty-five years in software, with a decade of platform work at Atlassian spanning policy, FedRAMP and FIPS compliance, cloud migrations and provisioning. Technical direction for distributed teams delivering to Google and LinkedIn. Four US patent filings and a credited open source contribution.

Languages · Kotlin, Java, Python, TypeScript, Go

Selected AI Agent Work

Atlassian, 2025

Production AI agent for support operations

Delivered an agent to replace a manual escalation loop in which support engineers had to coordinate with backend developers to get operational requests executed.

  • Support engineers could not hold direct backend API credentials, so the agent was designed as the execution authority, with permission and validation safeguards implemented in the agent codebase.
  • Encoded recurring support workflows as scenario prompts capturing both the resolution procedure and the APIs required, running discovery interviews with support engineers to ground each scenario in how the work was actually done.
  • Ran a continuous feedback loop through in-product feedback capture and a direct channel with the support team, validating releases with end users and refining scenarios from real usage.

Stack: Atlassian Forge, Rovo Agent modules, TypeScript.

Agent evaluation framework

Identified that no adequate tooling existed to test the agent continuously and catch regressions, and built a standalone evaluation framework to close the gap. Subject of a pending patent application (see Patents).

  • Autonomous multi-turn testing: Tests are written as a scenario and an expected outcome; a testing agent drives the conversation dynamically rather than requiring every turn to be scripted, and the completed conversation is evaluated against the intended outcome.
  • Hybrid evaluation: Combines LLM-as-judge qualitative assessment with deterministic assertions over execution traces, making behavioural quality and tool-call correctness both checkable.
  • Architecture: Separates orchestration and conversation from evaluation and observability, preserving the full conversation trajectory plus execution metadata: tool and plugin invocations, parameters, outputs, latency and token usage.
  • Generalised across domains: Applied the same ReAct architecture to evaluating coding agents, implemented independently in Python, showing the pattern held across a different agent domain and runtime.

Stack: LangGraph (TypeScript and Python), ReAct agents, LLM-as-judge evaluation.

Experience

Atlassian

Atlassian

2015 - Present · 11y 7m
Principal Engineer - Central AI, Search Relevance Platform
Working along ML engineers to improve search ranking and the quality measurement of 3P connectors. Automated e2e tests and ML pipelines to compute MRR and Recall metrics.
Principal Engineer - Enterprise Trust
Cloud Access and Intelligence - delivered the support agent and evaluation framework described above.
Policy Platform - turned around a programme facing delivery issues; main contact across multiple departments. Cut operational burden and simplified the architecture on Open Policy Agent, contributing new features upstream to the open-source project.
FedRAMP & FIPS Compliance - drove cross-departmental compliance strategy: service duplication and network boundaries, FIPS-approved encryption across tech stacks, cost-effective alignment between departments.
Principal Engineer - Cloud Migrations
Owned architecture, technical direction and operational maturity for the team that shipped the Jira Cloud Migration Assistant and its data export/import orchestration services.
Tech Lead - Cloud Provisioning and Orchestration
Owned on-demand provisioning microservices; version-controlled monitors and alerts, authored the team's Technical Credo, and built a service maturity dashboard to target quality investment.
Senior Software Engineer
Used proofs of concept and data to justify rewriting a critical service, leading the change to a substantially smaller codebase; adopted Sonar to track tech debt and test coverage.
Globant

Globant

2006 - 2015 · 9y 2m
DreamWorks AnimationSenior Software Engineer, R&D - DreamWorks Animation
Los Angeles. Member of the team that built the asset management system behind Apollo, DreamWorks' digital design and CG media platform. Introduced microservices, 12-factor apps, lightweight containers and continuous deployment on PaaS to the Java platform. Ran exploratory research projects across Apache Kafka, Cassandra and MongoDB.
Technical Director
Owned technical direction across an account portfolio in the US and Argentina: chose technologies and processes, coached technical leaders from architecture down to hands-on work, and partnered with program managers on account health and pre-sales.
Senior Project Manager
Coordination of teams focused on Java enterprise developments for clients such as Google, SAC Capital, LinkedIn, Nike.
LinkedInTechnical Project Manager - LinkedIn
Offshore product development in C++, Java and JavaScript; backlog grooming and release management with LinkedIn's product owner, plus escalation resolution.
GoogleProject Leader, Social Networks - Google
Led projects helping clients adopt OpenSocial, Shindig and Memcached.
Java EE Architect
Led onsite/offshore teams across international engagements: a margin-trading reporting system for SAC Capital (USA) where daily release timing was operationally load-bearing, call routing for Coolwave (Netherlands), and a hundreds-of-KLoC pharmaceutical research platform for Vertical-I (Switzerland).
Organization of American States

Organization of American States

2010

Software Development Consultant. Agile adoption and software configuration management; led ASP.NET/MSSQL web development and Sahi + TestNG test automation.

Technisys

Technisys

2003 - 2006

Java EE Architect & SQA Leader. Implemented CI/CD continuous, developed a central quality dashboard. Architect and lead developer for home banking sites.

Banco Galicia

Banco Galicia

1999 - 2000

Programmer and Data Modeler. Visual C, Visual Basic, ASP and COBOL-DB2 modules.

Patents

Methods, Systems, and Apparatuses for a Judge AI Serviceapplication filedU.S. Patent App. No. 19/407,202
Machine Resolution of Multi-Context AcronymsU.S. Patent No. 11,640,422 · 2023
Systems and Methods for Migrating DataU.S. Patent No. 11,500,834 · 2022
Systems and Methods for Project MigrationU.S. Patent No. 11,481,213 · 2022

Education

University of Belgrano - IT Engineer, Computer Science

University of Belgrano - IT Engineer, Computer Science

1998 - 2004

Grade 94/100. Thesis: an XACML-based web services authorization system, scored 10/10. Teaching assistant. Academic Council member.

Certifications

Amazon Web ServicesAdvanced Developing on AWS - Amazon Web Services2020
Amazon Web ServicesAWS Certified Solutions Architect, Associate2017-2020
Professional Scrum MasterProfessional Scrum Master2010
Project Management ProfessionalProject Management Professional (PMP) - Project Management Institute2009-2012

Open Source Contributions

Regal (open-policy-agent) · Go - Designed and implemented aggregate rules, the headline feature of v0.10.0 (PR #323), working the public API contract with the project maintainer.
@atlassian/danger-plugin-atlassian · TypeScript - Author. DangerJS plugin for Bitbucket Cloud, published to npm under Atlassian Labs (Apache 2.0). Provides idempotent PR task creation and reviewer assignment against the Bitbucket Cloud API.