Six years building software. Now focused on Business Intelligence, analytics engineering, and data quality for SaaS products. The transition was natural — I always cared more about what the data was saying than the screen displaying it.
I follow the data from source to decision: backend events, GA4 exports, tracking pipelines, BigQuery models, scheduled SQL transforms, Power BI semantic layers, and stakeholder-ready diagnostics.
Currently: building and maintaining the NoPing BI platform — raw backend/GA4 sources, noping_web analytical tables, dependency-ordered scheduled queries, a versioned Power BI semantic model, dashboards, funnel analysis, revenue/client cohorts, marketing performance, software usage, and CRM automation datasets.
| Impact | What I did |
|---|---|
| BI platform ownership | Maintain the NoPing analytics stack from raw backend/GA4 sources to BigQuery tbl_* tables, scheduled SQL, Power BI TMDL semantic model, and report pages |
| 19 active analytical tables | Organized the warehouse into domain tables for trials, sales, revenue, clients/churn, marketing, software performance, access type, page views, and devices |
| 6-page Power BI report | Built and documented a version-controlled .pbip report with Revenue, Trials, Sales, Clients, Software, and Marketing pages, backed by TMDL, JSON visuals, DAX measures, relationships, parameters, and bookmarks |
| Data warehouse cleanup | Migrated analytics tables from noping_analytics to noping_web, removed 20 orphaned tbl_* tables, deleted an unused latency schedule/table, and documented destructive BigQuery actions in changelog |
| CRM automation datasets | Built incremental email-segmentation tables for expired trials and monthly-plan users: played, logged but did not play, never opened the app, and did not renew |
| Visit → Register root cause | Proved a reported funnel collapse was a BI artifact, not a product regression — GA4 trial tracking stopped, the visits denominator moved to backend tracking, and real trials stayed stable at ~800-900/day |
| GA4 visibility gap | Cross-checked GA4 with backend events and found that traffic in GCC/Asia was still arriving while GA4 was being filtered or blocked; recommended backend as the source of truth |
| 215x backend/GA4 gap | Found China was almost invisible in GA4 (3 GA4 visits vs 645 backend visits/week), revealing real traffic hidden from dashboards |
| China conversion diagnosis | Investigated low CN registration conversion with BigQuery, VPN tests, backend review, and temporal analysis; identified Cloudflare Turnstile friction plus residential bot/proxy traffic as the main drivers |
| Tracking data quality | Mapped backend visit dedup (md5(ip+userAgent), 24h TTL), validated it empirically, and documented where bots, race conditions, prefetch, and missing user/session IDs still distort metrics |
| Timezone bug found | Discovered trial_page_visits.inserted_at was saved as BRT but interpreted as UTC by BigQuery, affecting temporal analysis and monitoring |
| 33x | Found a bug silently inflating core business metrics by 33x |
| 83 GB → incremental | Rewrote full-scan queries to incremental loads |
| 19h → 1h19min | Reduced BigQuery slot time by ~93% |
| Dependency-ordered ETL | Scheduled BigQuery refreshes across business domains with explicit dependencies, including revenue before clients/LTV and software overview before cross-usage |
| 61-month cohort retention | Built cohort curves tracking user retention from month 0 to month 60, segmented by acquisition plan and top game |
| End-to-end attribution | Last-touch model with 30-day window covering CAC, LTV, ROAS, CPT, CPC, CTR, CPM — from Google Ads spend to subscription revenue |
| Full ELT from desktop to cloud | Designed the entire pipeline: Electron app (TypeScript) → sandboxed IPC → main process enrichment → secured Cloud Run endpoint (Go) → BigQuery streaming insert. Events enriched in real-time with geolocation, OS fingerprinting, and hardware telemetry |
| Per-session hardware telemetry | Each gaming session records RTT, packet loss, bridge routing, CPU/GPU usage, CPU/GPU temps, and RAM — correlated with ping improvement and ISP data |
| 12-phase user funnel | Instrumented the full user journey across 9 BigQuery tables: boot → auth → connection → 8 feature screens, with session stitching via hardware fingerprint |
Data 
Cloud 
Dev 
