Led the data engineering execution of Meta's Safe Ads Program — a company-critical initiative transitioning advertising infrastructure from broad data access to consent-first architecture across four privacy domains (AAP, Consent Revocation, BYOU, ICOU). Landed 800+ diffs, remediated 700+ data assets, and discovered systemic platform-level validation gaps that prevented silent production failures at scale

• Discovered that the entire pre-deployment validation stack — access checks, schema validation, CI/CD gates — was blind to three distinct runtime privacy enforcement layers when pipelines transitioned from indirect (auto-projected) table references to direct cross-namespace reads during the consent migration
• Identified and documented 8 distinct failure classes spanning advertising data boundary restrictions, column-level user data policy blocks, export control violations, policy-zone access denials, signal-table misrouting, cross-namespace query engine incompatibilities, multi-consumer ACL gaps, and namespace reference errors
• Designed and deployed an automated boundary-guard system that probed all 191 in-flight migration diffs, flagged 54 at-risk cross-boundary swaps with hard-gate "do not land" conditions, and prevented further production incidents
• Implemented a runtime-simulation linter sweep as an authoritative pre-land gate — caught a real block that the existing metadata-based probe had falsely marked as passing
• Managed incident response for 5 production failures including 2 SEV3s (one with $5.27M/day revenue impact); executed fix-forward strategy with zero reverts by coordinating cross-team permission grants across 6+ consuming teams
• Handled director-level escalations with clear RCA documentation, established new operational protocols, and prevented recurrence across the remaining migration set
• Documented 6 CI false-positive patterns and 7 tooling issues that were causing wasted investigation time, false alarm pings to partner teams, and invisible landing blocks — turning tribal knowledge into repeatable triage procedures

• Co-designed with SWEs a custom AI coding skill (structured playbook) encoding the complete remediation methodology for all four privacy domains — enabling any engineer to execute complex multi-step pipeline migrations with AI assistance from day 2 (vs. ~2 weeks prior ramp-up)
• Skill covers end-to-end: table transform generation, consent filter pipeline creation, signal table dependency wiring, multi-producer alignment, lint/test automation, cross-team dependency resolution, and boundary-guard integration
• Cut per-asset remediation time by 3–5× and enabled a single engineer to sustain 800+ landed diffs — a throughput level typically requiring 3–4 engineers without AI augmentation
• Managed a 276-diff BYOU migration batch with 231 successful landings (84% first-pass success rate), triaging the remaining through permission grants, code fixes, and export control resolutions
• Collaborated with software engineers to continuously improve the AI implementation: expanded coverage from AAP-only to CR, BYOU, and ICOU domains; added cross-namespace boundary detection; eliminated fragile metadata tracking patterns
• Applied prompt engineering and LLM-assisted code analysis to reduce documentation overhead by ~70%, freeing engineering time for higher-judgment work (policy-gap analysis, cross-team grant negotiations)

• Architected an 8-stage remediation pipeline (classification → lineage → topology sort → pipeline dev → diff → deploy → compliance → downstream migration) serving as the execution framework for 700+ data asset remediations across AAP, CR, BYOU, and ICOU
• Achieved 100% remediation of all high-risk assets across AAP and Consent Revocation on schedule — directly unblocking Meta's Safe Ads regulatory launch
• Built consent filtering pipelines enforcing user-level opt-out at the data layer, signal table architectures for auditable dependency management, and differential privacy operators preserving analytical utility under consent constraints
• Navigated a 639-diff commandeered stack with deduplication challenges, namespace pollution, and non-deterministic version control states — establishing processes for safe editing of large-scale migration batches
• Closed compliance gaps across cross-functional pipelines involving 3+ producer teams — required resolving ambiguous ownership, negotiating permission grants that only consuming teams could submit, and coordinating export control reclassifications
• Reduced average asset remediation cycle time from weeks to days; cut time-to-first-deploy for new pipelines by ~50% through documented, repeatable validation workflows

• Designed prompt templates with stable cacheable prefixes + minimal variable context — cut input tokens 60% across 800+ AI-assisted diffs
• Routed simple transforms to local template engines (zero API cost); reserved Claude for ambiguous multi-consumer rewrites only — eliminated 30% of unnecessary invocations
• Enforced retrieval-over-stuffing: pass 40-line function blocks, not 2,000-line files. Structured JSON output schemas cap response length to what's needed
• Capped agent retry loops at 3 iterations with pre-flight input validation — killed retry spirals burning 40x normal token budget on stale inputs
• Tracked cost per landed diff (not per API call) — optimized for first-pass success rate, driving effective cost to ~$0.80/remediation at 800+ diff scale

Strategic architect for VMware's Partner Data Platform — unifying fragmented partner data into a single governed analytics layer

• Designed and built the Partner Data Platform from scratch, unifying 10+ disparate sources — eliminating data silos blocking cross-functional decisions for years
• Built data catalog and governance model (Python/Confluence) cutting tribal knowledge reliance by 60% and ad-hoc requests by ~40%
• Implemented VMware's first structured data governance for partner data — classification, lineage, policy enforcement — achieving audit-ready status
• Improved data consistency by 60% through standardized naming, automated quality checks, and cross-team contracts

• Pioneered an AI copilot for retrieval optimization and governance — one of VMware's earliest LLM-assisted data engineering deployments, reducing query dev time by ~30%
• Achieved 40%+ Spark processing time improvements through broadcast joins, caching, and skew management
• Established automated monitoring replacing reactive firefighting, reducing incident response from hours to minutes

• Drove alignment across 5+ data teams via shared roadmaps, reducing duplicate efforts by ~25%
• Led workshops upskilling 30+ partner data consumers, building organizational data literacy
• Engaged director/VP-level stakeholders, securing budget — team grew from 3 to 8 engineers

Led high-impact projects across Meta's Ads, Commerce, and Privacy teams

• Architected centralized DW unifying product/seller data — reducing decision-making cycle from weeks to days
• Launched Category Management Data Warehouse enabling cross-vertical analysis and measurable GMV growth
• SMB funnel redesign: 3x query performance improvement, scaling to tens of thousands of advertisers

• Led privacy remediation across 1,200+ SMB 2.0, Customer Journey, and BPO tables — establishing patterns adopted in Safe Ads
• Enhanced anonymization with Hive engineers, reducing audit preparation effort by ~60%
• Drove Salesforce ID deprecation across 1,000+ pipelines, eliminating vendor dependency affecting ~15% of joins

• Migrated 1,000+ Hive pipelines to Spark, automating ~70% — cut projected 12-month migration to under 6 months
• Built chargeback/leakage dashboards (FGF, Dataswarm, Unidash) identifying previously undetected revenue leakage
• Designed Marketplace App reliability dashboard — reducing MTTR for payment incidents via real-time visibility

• Migrated 1,000+ Hive pipelines to Spark; built chargeback representment & Marketplace reliability dashboards

• Led GDPR compliance: encrypted sensitive data into Dali storage using Hive, Python, and Pig; retired legacy sources

• 80+ Dataswarm pipelines for petabyte-scale ads: Cross Device Insights, Global Account Pipeline, Outcomes Datamart + Norms DB, Facebook Media & Live Monetization

• PII masking, Vertica→Redshift migration, 1,000+ table optimization

• Life Sciences BI, cloud migration, ETL modularization

• Incentive comp analytics, BusinessObjects XI

• Enterprise DW + 8 datamarts; ETL reduced from 18 hrs to 3 hrs; 40–50% sales lift via clickstream analytics