Blog
Practical writing about AI tools, datasets, document processing and the things I build.
Agents & assistants 2
- Building a Minecraft Style Browser Game with AstraA personal browser game experiment with one playable original, successive screenshots and notes on the Astra and Codex revisions.
- BotSim: A Local AI Agent Community SimulatorExplore scheduling, memory and restart behaviour in a small local agent community, with deterministic fixture checks kept separate from live model behaviour.
Applications & utilities 3
- Extracting Contact Fields with Source EvidenceTurn saved text into contact fields while keeping each value attached to its source passage and separating extraction from verification.
- Image Resizing: File Size, Formats and MetadataChoose image dimensions, fit rules and file size limits, then check the actual output and metadata instead of relying on format assumptions.
- Processing Saved Chat Logs Without Losing HistoryProcess saved chat JSON into traceable fields while preserving message IDs, edits, timestamps and source pages. This is an offline export workflow.
Datasets & documents 4
- Synthetic Datasets with Independent ValidationGenerate synthetic examples from checkable source facts, validate answers independently and split parent records before producing related variants.
- PDF to Markdown: Checking Extraction FidelityCheck whether PDF extraction preserves page evidence, reading order and meaning, and identify when native text extraction needs an OCR alternative.
- Organising Research Papers, Books and Reading NotesKeep selection, acquisition and reading separate in a research collection, with source versions and the question behind each saved paper or book.
- Dataset Cleaning Without Losing IDs or MeaningClean datasets with explicit row accounting, preserved identifiers, conflict quarantine and reproducible splits rather than silent type conversion.
Developer workflows 1
Knowledge & memory 4
- Xanadu Inspired Links to Versioned Document PassagesKeep document versions, passage spans and link identities intact through export and import, including Unicode text and historical source references.
- Organising Research Notes Without Losing HistoryCreate a useful reading copy of research notes while preserving original files, duplicate relationships, conflicting versions and attribution.
- Building Technical Knowledge Bases for TroubleshootingOrganise a technical knowledge base around the next useful check, with source evidence, device time and receipt time kept distinct.
- Embedding Compatibility: Models, Vectors and SearchRecord model revisions, pooling and normalisation beside embeddings, then compare retrieval methods without mistaking equal dimensions for compatibility.