Junior Data Analyst · Middle East Historian · Brandenburg, Germany
I work with messy, multilingual and fast-changing data – from Persian-language Telegram channels to on-chain transactions – and I document how every result was reached, including its limits. My background is in Iranian, Islamic and Jewish studies. Python, SQL and Power BI are how I apply a historian's source criticism to data.
- Source criticism: I check where data comes from and what it can and cannot show. I state uncertainty instead of making results look stronger than they are.
- Regional and language knowledge: Iran and the wider Middle East; Persian (native), German (fluent).
- End-to-end work: collecting data (APIs), cleaning and validating it, modelling it (star schema, PostgreSQL), analysing it and presenting it (Streamlit, Power BI, Dune dashboards).
| Project | What it does | Live |
|---|---|---|
| osint-berlin-2026 | Berlin public safety picture, Jan to Oct 2026, from open sources: 1,272 Berlin police press releases (scraped, gaps filled via Wayback Machine and Common Crawl) compared with 8,807 media articles (GDELT/BigQuery); categories by a locally run language model, districts alongside population, income and employment data; bilingual dashboard | Dashboard |
| persian-media-analysis | 328,000+ Persian-language Telegram posts from six Iranian news channels during the 2026 war: PostgreSQL database, locally run AI classification, bilingual dashboard | Dashboard |
| iran-sanctions-analysis | All 62 recorded sanction cases against Iran, 1979–2025, compared with 1,794 cases worldwide (Global Sanctions Data Base): who imposed them, with which instruments, why and with what result; plus 2,840 sanctioned Iranian companies, persons and vessels as of 2026 by actor, sector and office (OpenSanctions); bilingual dashboard | Dashboard |
| neurodiversity-lgbtq-german-news | How often four German news websites name ADHD, autism, neurodiversity and LGBTQ+ in their headlines, Nov 2024–Sep 2026: 188,354 text items from tagesschau.de, SPIEGEL, WELT and ZDFheute; reason for LGBTQ+ coverage classified by a cloud and a local language model; bilingual dashboard | Dashboard |
| cybercrime-germany-pks | Recorded cybercrime in Germany, 2016–2025: cases, clearance rate and differences between the 16 federal states, from the BKA Police Crime Statistics | Dashboard |
| right-wing-crime-germany-pmk | Right-wing motivated offences in Germany, 2014–2025: offence types, differences between the 16 federal states and AfD election results alongside, from official BKA/BMI statistics | Dashboard |
| rss-news-automation | Fully automated daily email digest with news from five countries: an n8n workflow collects RSS feeds, stores them in PostgreSQL, draws a random sample and has a local AI model (Ollama) summarise it; a further stage compares these summaries with a cloud model | – |
These projects are not about making money with trading. I love Web3 because I see a real future for it in the real economy. It offers maximum transparency: users can follow what actually happens to the money they invest somewhere. Not everything, but much more than with traditional banking. Everything stays in one place and is traceable. It also allows independence from many intermediaries, and crypto could become simple to use in everyday life. I'm also interested in new ways of buying and selling stocks, such as tokenized shares, once the legal situation improves. And of course in strange phenomena like the destructive buying and selling of memecoins, for no real reason at all.
| Project | What it does | Live |
|---|---|---|
| jupiter-perps-wallet-tracker | Risk monitoring for a Solana wallet on Jupiter Perps, not a trading bot: it never executes trades. Telegram alerts based on fixed risk rules, nightly report with charts and a local AI model (Ollama), backtest and paper trading; Python service orchestrated by n8n | – |
| solana-jupiter-trading-analysis | SQL on 465M Solana swap events routed through Jupiter, H1 2026 | Dune dashboard |
| dune-api-pipeline | ETL pipeline: Dune API → Python (clean, validate) → star schema → seven-page Power BI dashboard | – |
| ethereum-lending-analysis | Aave, Compound, Morpho and Spark, Jan–Aug 2026: borrowing, users, asset mix and liquidations (Dune SQL) | Dune dashboard |
| usdc-ethereum-solana-analysis | How native USDC was used on Ethereum and Solana in H1 2026 (Dune SQL) | Dune dashboard |
| tokenized-vs-nasdaq-stocks-after-hours | Five tokenized stocks on Solana vs. their Nasdaq shares, Jul 2025–Sep 2026: what the token price does while the exchange is closed | Dashboard |
| multi-asset-dca-portfolio-analysis | A real multi-asset portfolio and a simulated crypto DCA strategy, May 2025–May 2026: performance, risk, benchmarks, dimensional data model | – |
- global-defense-sector-analysis – 42 defense companies from 15 countries, 2022–2026 (Python, Power BI)
- car-industry-germany-vs-asia – Germany vs. China, Japan and South Korea, 2019–2025 (Dashboard)
- german-credit-rates-vs-ecb-rate – German credit rates vs. ECB policy rates, 2020–2026 (Dashboard)
- dax-mdax-h1-2026-analysis – DAX vs. MDAX in the first half of 2026
- covid-market-sector-analysis – 24 assets in four sectors before, during and after COVID-19 (2018–2023)
- follow-for-follow-checker – Tracks views, visitors and clones of my GitHub repos and checks who follows back (Python, GitHub API, Excel)
Python (pandas, Matplotlib, Seaborn, Streamlit) · SQL (PostgreSQL, Dune) · Power BI (DAX) · Excel (Power Query) · APIs & ETL · Workflow automation (n8n) · NLP and local language models (Ollama) · AWS basics · Jupyter · Linux
My working environment as a NixOS configuration: nixos-data-analysis. My daily system is Arch Linux, but this file contains the same tools I work with.
I'm an analyst, not a software engineer. Many of my projects were built with AI assistance, using cloud and local models. I define the question and the method, check every result myself and take responsibility for it:
AI can speed up work. It does not take responsibility for whether a result is correct.
How I work with AI, which models I tested and where they went wrong: How I Actually Use Local and Cloud AI.