Joyan Bhathena Joyan9

Hi there, I'm Joyan 👋

Welcome to my GitHub profile!

Data Analyst with a Master’s in Computer Science and hands-on experience transforming complex data into business impact
At DataVinci, built automated dashboards and data pipelines that:
- Increased client ROAS by 20%
- Significantly reduced infrastructure costs
At AUTO1 Group, bridged business and technical teams to:
- Automate workflows
- Streamline operational processes
Currently seeking full-time opportunities as a Data Analyst or Analytics Engineer

Outcome: Prevented €390K/month in potential revenue loss by recommending against a product page redesign that looked like a win on engagement (+82% time on page) but caused a -15.7% drop in revenue per session. Identified a salvageable mobile opportunity worth ~€10K/month.
Approach: Analysed 187,974 sessions and 8,572 conversions across a 4-week A/B test. Started with overall metrics, then segmented by device, user type, product category, and price range to uncover the real story hidden in the averages. Applied chi-square testing (99.9% confidence) and quantified the business impact in €, not just percentages. Delivered a 1-page strategic memo to stakeholders.

Outcome: Identified two independent root causes behind an unexpected February revenue drop — a mobile add-to-cart failure and a CPC attribution breakdown — with a combined impact of ~$11,000 in lost revenue. Delivered a prioritised fix plan with concrete next steps for both engineering and marketing.
Approach: Worked directly from the raw GA4 BigQuery export (~113K rows, 61 days) with no access to the GA4 UI. Established a pre-anomaly baseline using 7-day rolling averages, broke down traffic by channel to detect attribution shifts, built a full session-level funnel (session_start → purchase) for both periods, and cut it by device to isolate the mobile-specific failure.

Outcome: Pinpointed the root cause of a 31% delivery time degradation (9 → 12+ minutes) across a quick-commerce platform and delivered a 1-page memo to the COO & VP of Product — with data-backed arguments against three competing executive hypotheses (seasonal rain, marketing pause, mass rider hiring).
Approach: Worked across three datasets (109K orders, 34K store snapshots, 7 external events) to test each hypothesis independently. Used SQL window functions and JOINs to correlate store-level rider availability, queue depth, and external events with delivery time spikes across zones and time periods.

Outcome: A fully automated daily pipeline that fetches, deduplicates, and AI-scores job listings across Germany, writing only pre-filtered results to a Google Sheet each morning. Eliminated the need to manually scan job boards, reducing daily job search overhead to a 5-minute review.
Approach: Built a multi-stage pipeline: JSearch API for aggregated listings → SHA-1 fingerprint deduplication and stale-posting filtering → Groq LLM consensus scoring (two models, structured Pydantic output, chain-of-thought reasoning) → Google Sheets write via service account. Deployed on GitHub Actions (cron 06:30 CET) with git-based state persistence. Designed to be model-agnostic and configurable via YAML.