How to Evaluate PDF Extraction Accuracy with Python
Evaluate PDF extraction against reviewed labels using field-level precision, recall, document completeness, and a reproducible Python scoring example.
Blog
Technical notes and research insights.
Short articles on AI engineering, document processing, knowledge graphs, research software, open-source work, and building dependable systems.
Idempotent Python Scheduled Jobs: Safe Retries and Recovery
Design scheduled Python workflows that can run again safely, with durable run keys, transactional results, bounded retries, and explicit recovery steps.
Incremental API Sync: Cursors, Checkpoints and Safe Restarts
Build resumable API synchronization with atomic page checkpoints, version-aware upserts, deletion handling, cursor recovery, and reconciliation.
SHACL Validation in Python: Check RDF Data Before Publishing
Validate RDF data with SHACL and Python using a runnable product example, explicit datatype rules, useful reports, and a staged publishing workflow.
Webhook Idempotency: A Durable Inbox for Duplicate Events
Handle duplicate webhooks with a durable inbox, atomic event keys, safe worker retries, and explicit boundaries for database and external side effects.
How to Use Google Search Console to Decide What Content to Improve Next
A practical workflow for finding SEO opportunities from Search Console impressions, CTR, average position, pages, and query intent.
Knowledge Graph vs Relational Database: When a Hybrid Model Helps
Compare knowledge graphs and relational databases for multi-source data, identity, provenance, validation, and queries with a practical decision table.
How to Monitor a Web Scraping Pipeline Without Checking Data by Hand
How to detect scraper breakage with data-quality checks, schema monitoring, anomaly detection, crawl metrics, and targeted alerts.
How to Debug a Multilingual Sitemap When Google Is Missing Pages
A practical checklist for multilingual sitemap, canonical, hreflang, routing, and Search Console issues when Google discovers only part of a website.
How to Keep API Keys and Admin Secrets Safe in a Frontend + Backend App
A practical deployment pattern for keeping API keys, database credentials, and admin secrets out of public frontend code.
How to Send Transactional Email When a Hosting Platform Blocks SMTP
Practical alternatives for sending transactional emails from deployed applications when outbound SMTP is restricted or unreliable.
SQLite to PostgreSQL Migration: Plan, Validate and Cut Over
Plan a SQLite to PostgreSQL migration with type checks, repeatable imports, sequence repair, verification, and a cutover plan that accounts for new writes.
How to Connect Business Systems with APIs, Webhooks and Reliable Data Sync
Connect business systems through APIs and webhooks with explicit data ownership, retry-safe operations, reconciliation, useful logs, and protected credentials.
How to Turn Company PDFs Into a Searchable RAG Knowledge Assistant
A practical RAG architecture for turning internal PDFs into searchable knowledge with OCR, chunking, retrieval, citations, validation, and access controls.
Automated Website Reports by Email: Audit, Schedule and Deliver
Build automated website reports by email with scheduled audits, useful change summaries, retry-safe delivery, and clear handling of incomplete checks.
Building Before You Trust: Evidence-First Public-Web Search
How I built Before You Trust with SearXNG, YaCy, identity confirmation, source filtering, privacy controls, and evidence-first search.
JSONL to PostgreSQL: Designing a Reproducible Product Catalog Pipeline
How to use JSONL stage files, object storage, validation, idempotent loading, and PostgreSQL to build a reproducible product catalog data pipeline.
Reliable E-commerce Web Scraping: Locale Redirects, Schema Drift and Validation
Lessons from scraping e-commerce sites reliably when locale redirects, changing HTML, incomplete categories, anti-bot behavior, and schema drift break assumptions.
Product Matching Across Stores and Languages: E-commerce Entity Resolution
How to match the same product across e-commerce stores and languages using identifiers, normalized attributes, variants, and conservative entity-resolution rules.
E-commerce Data Pipeline Design: Scrape, Normalize, Enrich, Validate and Load
A practical e-commerce data pipeline architecture for scraping product data, normalization, enrichment, validation, product matching, and database loading.
Building Orgelia: Architecture for a Multi-Country Product Comparison Platform
How I separated frontend, API, catalog data, shop discovery, maps, and data ingestion while building Orgelia as a multi-country product comparison platform.
Website Information Architecture for Small Businesses: Lessons from Websiteli
What I learned simplifying Websiteli's navigation, service positioning, portfolio, pricing, and conversion paths for a small-business technology website.
Building Websiteli with Astro: Multilingual SEO for a Service Website
How I structured Websiteli with Astro, localized routes, canonical URLs, hreflang, structured data, sitemaps, analytics, and reusable SEO content.
Static Website Form to Google Sheets with Google Apps Script
How to connect an Astro or static website form to Google Sheets using Google Apps Script, with validation, UTM tracking, email status, and no backend server.
GitHub Pages Custom Domain and HTTPS: Lessons from Slap AI
A practical guide to GitHub Pages custom domains, DNS records, HTTPS certificate issues, apex vs www, and deployment debugging based on Slap AI.
Building Slap AI: An Astro Pre-Launch Site on GitHub Pages
How I built Slap AI's pre-launch website with Astro, GitHub Pages, conversion-focused CTAs, analytics consent, and a lightweight static architecture.
How to Extract PDF Sections Reliably When Headings Are Missing
A robust strategy for PDF section extraction when expected headings are missing, reordered, duplicated, or detected imperfectly.
PDF Document Processing in Python: OCR, Extraction, Validation and RAG
Build a Python PDF processing pipeline with native text extraction, page-level OCR, layout evidence, field validation, and a deliberate human-review path.
Software Taxonomy for Research Software: Why Classification Is Hard
Understand software taxonomy for research repositories: separate functional and scientific categories, compare metadata, and evaluate classification clearly.
Building an ontology for a sensitive product domain
Why the Sex Toy Ontology is an interesting test of inclusive modeling, product safety, and useful semantic infrastructure.
What documenting GSoC taught me about open-source work
Lessons from recording the Ontology Time Machine project week by week instead of showing only the final result.
LlamaIndex PDF Querying: Where It Fits in a Reliable Document Pipeline
Use LlamaIndex for PDF querying with a clear parsing boundary, structured metadata, source evidence, and evaluation of retrieval, answers, and citations.
Documentation is operational resilience
Clean code and concise documentation reduce single-person dependencies and make systems easier to maintain.
Zero-shot learning, explained simply
How models make useful predictions for tasks and categories they were never explicitly trained to recognize.
AI · data engineering · applied research
Discuss an AI, data engineering, or research collaboration.
Reach out about roles, consulting conversations, research questions, or technical collaboration in Zurich, Switzerland, or remote teams.