Volunteer data pipeline for Greenpeace Israel's campaign team. Built a full ETL pipeline pulling Google and Meta Ads data via Coupler.io → Python transformation scheduled on GitHub Actions → Google Sheets via authenticated API → Looker Studio dashboard. Enables cross-platform campaign performance and costs analysis for better marketing decisions.
A Bayesian A/B analysis of the Olist Brazilian e-commerce dataset (95,559 customers, R$15.9M spend) asking whether voucher-acquired customers are worth it. Split first-order cohorts into control, partial-voucher, and full-voucher groups, then ran a Dirichlet-weighted Bayesian bootstrap (10,000 iterations per group). The 180-day revenue verdict is decisive: control out-earns both voucher groups with 100% probability (~R$165 vs R$0.56 for fully-covered customers). The twist — full-voucher customers repeat the most (3.9% vs 3.0%) — resolves to a clear recommendation: full-cover vouchers buy visits, not revenue, and belong in the retention budget, not growth.
Multi-agent data analysis platform that covers DIG framework — a paradigm of data analysis using LLMs. Upload any data file → a hardcoded pandas layer extracts facts → 5 specialized LLM agents interpret and analyze along with the user → Consolidate Living Report with real insights built along the progress. Key decisions made: to control LLMs hallucinations pandas and plotly doing all the math (LLM never sees raw data), 'Pydantic v2' validates every agent output, parametric orchestration keeps Python in control of flow.
Autonomous AI integrated job scanner running in production. Monitors Telegram job groups 3× daily, scores matches against my DA profile using GPT-4o mini, scoring storage in Supabase for deep analysis and improvements. Telegram alerts for high-fit roles. Anti-hallucination design: temperature=0, Pydantic validation, confidence scoring tracking. The pipeline is completely adjustable for different candidates — see the manual and README of "JobPulse AI Scanner" on GitHub.
I'm making a deliberate shift into data. Driven by a want to work close to real problems, understand how things actually work, and build infrastructure that helps teams make better decisions. I care about the full stack - from a clean SQL cohort to the Telegram alert firing and UI. Currently i'm building data pipelines and AI-native workflows: an autonomous job scanner in production, a multi-agent analytics platform, and a volunteer data pipeline for Greenpeace Israel. Equipped with a B.A. in Statistics & Psychology (BGU, GPA 90) and the experience of running a small business, I bring a high degree of responsibility, adaptability, and ownership.
