Web & product
Metal Globe
I put 195,679 metal bands on a 3D globe, filterable by country, genre and decade.
three.js · 4,408 → 496 lines, split into data.js · lazy-loaded flags · static
Case study
research software engineer · lima · remote
Code as craft. Data as evidence.
I build data pipelines, scrapers and web tools for research and business, and I document the decisions so anyone can audit them.
16 projects · 8 live
Projects
Case studies with the source and the numbers behind them.
Science & bioinformatics
Atlas Botánico del Perú
I turned raw GBIF data into a tree of Peru's 21,585 accepted plant species that opens in the browser.
DuckDB · 1.28M GBIF records · ten-stage ETL · static marts, no backend
- DuckDB
- ETL
- Astro
- Apache ECharts
- Python
- +2
Case study
Machine learning & AI
Predictor de Ingresos e Informalidad
I compared 9 model specifications to predict labor income and informality risk in Peru, from the ENAHO household survey.
scikit-learn · nine-model tournament · statsmodels intervals · Streamlit
- scikit-learn
- pandas
- Streamlit
- statsmodels
- Python
Case study
Being moved to public
OpenMetadata Warehouse
In development · PrototypeInfrastructure & servers
OpenMetadata Warehouse
I connected Claude Code to an OpenMetadata service running over an Oracle and SQL Server warehouse, so ingestion, lineage and server lifecycle are managed as code.
OpenMetadata · Oracle · SQL Server · Python · Docker
- Oracle
- SQL Server
- Python
- Docker
- +1
Case study
Open Data Repository
In development · PrototypeData & public sector
Open Data Repository
I'm turning a folder of scraping and cleaning scripts into a static catalogue anyone can search, preview and use offline.
Python · SQLite · static site, no server
Workboard
In development · PrototypeWeb & product
Workboard
A Jira-style board: projects, epics, cards, timeline — built for a public-sector IT office.
JavaScript
Case study
Signal Agent
In development · PrototypeAgents & AI automation
Signal Agent
A market monitor where the model has to write a falsifiable prediction before it can be scored — 3 open, 0 resolved, the first ones due January 2027.
Python · GitHub Actions · SQLite
Case study
54 tools · picked per problem
Stack
The tools I work with, grouped by area. Versatility isn't piling up frameworks — it's picking the right one for each problem.
Languages 06
- Python 11 projects The general-purpose language for data, scripting and backends.
- JavaScript 7 projects The language that runs in the browser: interactivity without depending on a backend.
- Bash 3 projects The shell language: automating, chaining commands and moving files without opening an editor.
- TypeScript 2 projects JavaScript with types: errors that show up while typing, not in production.
- HTML & CSS 2 projects The base of any web interface, before any framework enters the picture.
- SQL 1 project The language for querying and modelling relational data.
Frameworks 07
- Astro 2 projects Generates static sites that ship zero JavaScript except where explicitly opted in.
- Vite 2 projects Build tool and dev server, fast even in projects with many modules.
- React 1 project Component-based UI library; the base of almost any modern interactive frontend.
- Streamlit 1 project Turns a Python script into an interactive web app without writing frontend code.
- Zustand 1 project Minimal state management in React, without heavier alternatives' boilerplate.
- Tailwind CSS Utility-class CSS: designing without jumping between stylesheet files.
- PyQt6 Desktop interfaces in Python, used when the app cannot live in a browser.
Data 12
- SQLite 4 projects Relational database embedded in a single file, no server to administer.
- Apache ECharts 3 projects Interactive charts for dashboards with real data volume.
- Three.js 2 projects 3D in the browser over WebGL, for visualisations and interactive scenes.
- DuckDB 1 project Embedded SQL engine for analysing large files (Parquet, CSV) without standing up a server.
- ETL 1 project Extracting, transforming and loading data from a raw source into something queryable.
- pandas 1 project Table manipulation in Python: cleaning, grouping and transforming data before analysis.
- Oracle 1 project Relational database used in large, long-lived enterprise systems.
- SQL Server 1 project Microsoft's relational database, common in Windows-based enterprise systems.
- Parquet Columnar, compressed file format built for analytics over large volumes.
- Polars A pandas alternative written in Rust, much faster on large datasets.
- Excel & reporting Reports and spreadsheets when the final audience lives in Excel, not a dashboard.
- MySQL Open-source relational database, the most common one on shared hosting.
Infrastructure & servers 05
- Git 3 projects Version control: history, branches, and the base of any code collaboration workflow.
- Static hosting 2 projects Serving sites with no backend: faster, cheaper, smaller attack surface.
- Data catalog: lineage, ownership and discovery across an organisation's sources.
- Docker 1 project Packages an app with everything it needs to run the same on any machine.
- Airflow Orchestrates data pipelines as task graphs with dependencies and retries.
Operating systems 04
- Windows 2 projects Desktop operating system, the most common development and verification environment.
- macOS 2 projects Building and verifying on Apple Silicon, Apple's current hardware architecture.
- Ubuntu Server 1 project The server edition of Ubuntu, a Linux distribution.
- Linux 1 project The operating system behind almost everything deployed on a server.
Security 02
- Secrets management 1 project Credentials out of the repo, always via environment variable, audited before publishing.
- Data privacy Synthetic or public data in demos: no identifiable real person.
AI & machine learning 09
- LLMs & agents 3 projects Language models orchestrated as agents, with evaluation built into the flow, not the pitch.
- scikit-learn 1 project Classic machine learning models: classification, regression and clustering.
- statsmodels 1 project Classic statistical models in Python: regressions with inference, not just prediction.
- PyTorch Machine learning framework for training and running deep learning models.
- OpenCV Computer vision library: detection, transformation and image processing.
- InsightFace Open-source face detection and recognition, handled carefully since it is biometric data.
- OCR & documents Turning scanned documents or images into text that can be searched and processed.
- Segment Anything (SAM) Meta's image segmentation model: cuts out any object with no prior training.
- Speech-to-text Transcribing audio to text to search, index or caption spoken content.
Scraping & automation 06
- Scraping 3 projects Extracting data from undocumented APIs or HTML, always from legitimate sources.
- GitHub Actions 3 projects CI/CD hosted in the same repo: build, test and deploy on every push.
- Playwright 2 projects Automates a real browser to test sites or extract data that needs JavaScript.
- Entity matching 1 project Deciding whether two different records describe the same entity, by text, code or image.
- BeautifulSoup & selectolax 1 project HTML parsers for pulling data out of pages that have no API.
- Selenium Browser automation for testing or scraping when a site needs real interaction.
Business 03
- Open data 2 projects Building with public, citable sources instead of client data.
- Data visualization 1 project Making a huge catalogue understandable in 30 seconds: the discipline of showing data.
- E-commerce 1 project Catalogue, cart and checkout: the mechanics of a shop, not just the storefront.
the reasoning, not the pitch
How I work
IchiSieben is the working name of Yoichi Palacios Tanaka. I build data pipelines, scrapers and web tools — the kind of software that reads a source, cleans it, and turns it into something a person can act on.
Every project here ships with the decisions behind it written down: why this database and not another, what the data actually shows, what I would change next. That's not a courtesy — it's how the work gets checked, by me or by anyone else.
I use AI where it earns its place: to move faster through boilerplate, to explore a dataset, to draft the first pass. The judgement on what to build and whether it's right stays mine, and the README says so.
one email, no form
Contact
A project, a role, or a stack question? Pick what fits and I'll get the right context first.