AgenticOps Welcome + vibe-coding-team-init + Skill Connectors · Product Ownership, Operations & AI-assisted Delivery

A sustainable operating model connecting AI project setup, tool release and reusable enterprise capabilities.

• Combined vibe-coding-team-init with AgenticOps Welcome to standardise project setup and connect registration, review, deployment, feedback and operations.
• Supports a team of 40+ people and the deployment, release and ongoing updates of 20+ vibe-coded tools.
• Defined release controls that freeze the reviewed version, verify code and configuration together, check runtime status and roll back failed updates.
• Expanded the platform through real use: environment-aware search, co-developers, screenshot feedback, notifications, status tracking, test-to-production release, backup and restore.
• Contributed user guides, installation flows, environment boundaries and safety guidance to Skill Connectors, including a Splunk content-management Connector.
• Use 13 internal Connector categories across Codex, Claude Code, GitHub Copilot and Trae to connect knowledge, Splunk content and delivery systems for everyday work.
• Turned repetitive steps and tacit know-how from data validation, content publishing, ticket handling and work reviews into a reusable personal CLI and Skill toolkit with built-in operating rules, permission boundaries, human confirmation and outcome verification. The Splunk Skill supports read-only searches and Report / Lookup / Dashboard management, using environment checks, change previews and post-action validation to reduce production risk; other Skills cover Confluence publishing, Jira issues and worklogs, work journals, questionnaire dashboards and product-decision synthesis.

Product WorkflowPlatform OperationsCLI / SkillsSkill ConnectorsAI Collaboration

Management Methodology · Agent Product Design · Delivery

A management insight workflow from trusted data and analysis to evidence-backed action—not just another Q&A bot.

• Helped define a Dashboard → Insights → Ask Omi → Actions → Reports loop, using Ontology to align time, entities, KPIs, rules and evidence; the current phase focuses on Find, Ask, Read and Explain.
• Initially built a management knowledge Agent with Copilot Studio + Teams for Dashboard/KPI discovery, metric explanation, access requests and Service Request recommendations.
• Defined a minimum Incident ontology covering KPI formulae, units, time fields, sample boundaries, entities, relationships, field mappings, rules and version governance.
• Advanced a Splunk-native MVP that combines dashboards, insights, conversational analysis and an ontology relationship view while inheriting the current user's data permissions.
• Designed an explanation flow that verifies the premise, compares periods, locates contributing dimensions, links representative Incidents and produces evidence-backed attention points without claiming correlation as causation.
• Defined an Agent evaluation methodology across Outcome, Trajectory, Interaction and Production, using deterministic graders wherever constraints are verifiable and calibrated LLM judges only for subjective quality.
• Translated the methodology into an evaluation platform v0.1 with Agent onboarding checks, versioned datasets and suites, isolated sessions, deterministic graders, batch trials and release gates; repeated trials, paired baselines and human decisions fed back into the golden dataset reveal real regressions.
• Compared Copilot Studio and Splunk-native approaches and validated real data, no-data behaviour and premise correction with 15 fixed questions and seven multi-turn scenarios.

Management InsightOntologyAgent EvaluationGolden DatasetRelease Gates

LLM Application Development & Long-term Iteration

Production enterprise metric Q&A service with 287 personal commits over nearly two years.

• Integrated enterprise LLMs and knowledge retrieval, supported follow-up parameters and history summaries, and rendered results as tables and charts.
• Added deterministic matching against valid business dimension values to reduce model errors in application and organisation names.
• Maintained scheduled refreshes for recommended questions, general guidance and metric knowledge.

PythonFlaskSQLData Q&A

Requirements · AI Workflow Design · Product Iteration

Connected an internal visual AI orchestration platform to production logs in OpenSearch.

• Built the application on an internal platform similar in form to Dify/Langflow and connected it to OpenSearch.
• Designed and launched AI Query, AI Trace and AI Explain for natural-language search, call-chain reconstruction and exception explanation.
• Used interviews to identify real issues around mixed logs, cross-machine fragmentation, incomplete asynchronous traces and missing explanatory context.
• Converted a manual log-onboarding SOP into a trackable flow covering configuration generation, human review, deployment and retry; used AI coding to accelerate implementation while owning requirements, workflow design, result validation and acceptance.

Enterprise AI PlatformOpenSearchUser ResearchWorkflow Design

Splunk · Data Governance · Rule Definitions · Delivery & Service Operations

• Contributed to a Splunk-based automated application access-control service, owning account and entitlement data onboarding, SPL rule definitions, reporting iterations and application delivery. Daily rules detect risks such as orphaned accounts, incomplete approvals, invalid approvers and delayed suspension of leavers.
• Helped establish a closed loop from daily checks and risk detection through weekly alerts, business self-assessment, remediation tracking and closure, reducing the checking cycle to within one day. As of August 2026, the team service had onboarded 155 systems, with 149 actively scheduled, nine core IAM controls and 905,582 account and entitlement records scanned.
• Delivered eight rounds of access-control and operational-compliance self-assessment, owning form rules, data validation, Splunk analysis and result reporting. The latest round covered five domains, 29 business units, 32 controls and 927 systems with 100% completion; its dashboard recorded 8,605 visits from 695 users.
• Built a compliance Q&A assistant that turns IAM control guidance, self-assessment result explanations and recurring manual enquiries into a governed knowledge service. Completed a Microsoft Teams Bot pilot and launched a web assistant for operations staff supporting the 927 systems in scope.
• Owned subscription onboarding, usage records and quality assurance for a disabled-AD-account service, supplying the previous 30 days of disabled-account data for timely leaver suspension and valid-AD checks. As of August 2026, it supported 132 systems and 546,787 calls and was included in the standard IAM solution.
• Audited 23,010 records published over one year and, within the defined source and account-type scope, covered all 21,964 target disabled accounts with publication within 48 hours.
• Contributed to UAR data-service migration and designed an approval-email processing flow; used AI-assisted implementation while owning requirements, workflow design, data validation and acceptance.

SplunkSPLIAM / UARCompliance Q&AData ValidationCompliance Workflow

Data Engineering & Anomaly Detection

A hand-built platform that learns historical patterns and continuously detects anomalies, shifting data quality from reactive investigation to proactive control.

• Designed Z-Score checks for null-rate deviation, DBSCAN for category-share shifts, IQR for mapping anomalies and FP-Growth for field-value associations; contributed 220 personal commits.
• Built a closed loop from data update and automated scan to alert, JIRA/ITSM ticket, human validation and rule feedback.
• In September 2025, detected product misclassification affecting 5,210 inventory units; another alert found duplicated Sell-in reporting within about 20 minutes, affecting 6,642 servers worth USD 17.9M, and the issue was fixed the same day.
• Reworked sampling, dynamic thresholds and baseline updates to address false positives, contaminated history and large-table extraction limits.

PythonSplunk SPLZ-ScoreDBSCANFP-GrowthData Governance

Splunk MVP · Product & Governance Workflow · AI-assisted Delivery

A governed loop in which deterministic rules, AI and people make the decisions each is best suited for.

• Designed a case lifecycle of rule screening → AI second review → human decision → remediation/exemption → closure approval, keeping high-risk decisions with people.
• Designed a control-authoring lifecycle in which AI drafts data requirements, read-only SPL and Reports, followed by human revision, approval, disabled publication, Dry Run and administrator activation.
• Unified audit trails for controls, rule runs, findings, evidence, case versions, AI conclusions, human actions, notifications and recurrence, with Reviewer/Responder/Approver separation.
• Delivered an installable Splunk App, configuration and case workbenches, operations dashboard, release package, deployment guide and illustrated user manual.

Splunk AppHuman-in-the-loopAudit TrailRemediation Loop

MVP pending real-system integration; I own problem definition, workflow and role boundaries, task decomposition, acceptance and delivery documentation.