personal blog, research, writing and ideas
This is where I keep the things I've written up and made ready to share. All my stuff is open knowledge, so feel free to read through and use it however you see fit <3
Mostly AI, security, tooling + ideas & things I've been building.
I also use this space to work through research questions, explore weird ideas and occasionally wander off into abstract tangents :)
Blog · Research · Running the code
Practical tools, experiments and the questions behind them. Open the blog index.
- Processing Saved Chat Logs Without Losing History
- Image Resizing: File Size, Formats and Metadata
- Extracting Contact Fields with Source Evidence
- Dataset Cleaning Without Losing IDs or Meaning
- Organising Research Papers, Books and Reading Notes
- PDF to Markdown: Checking Extraction Fidelity
- Synthetic Datasets with Independent Validation
- Embedding Compatibility: Models, Vectors and Search
- Building Technical Knowledge Bases for Troubleshooting
- Organising Research Notes Without Losing History
- Xanadu Inspired Links to Versioned Document Passages
Model mechanisms, agent systems, knowledge tools, numerical methods and security studies. Articles distinguish proposed experiments, tested components and observed pilot results. Open the research index.
- Autoresearch: Citation Tracking for Research Agents
- Local AI Memory: Retrieval and Temporal Validity
- Mixture of Perspectives: Decision Support with Disagreement
- Multi Agent Orchestration: Streaming, Failures and Replay
- Swarm Sequence Generation: Neighbour Coupling Tests
- CloudVec: Vector Index Freshness with a SQLite Outbox
- Graph Based Research Memory with Source Provenance
- OpenAbstract: Versioned Claims and Corrections
- SL5: Thread Aware Retrieval and Quote Attribution
- Protocol Resource Limits and Amplification Modelling
- Synthetic Signal Detection Under Distribution Shift
- Defensive Security Research Projects
- Glow Worm: A Small Byte Level Language Model
- LLM Self Correction: Evaluation and Stopping
- Self Rewarding Training: DPO and Evaluation
- LLM Arithmetic: Digits vs Number Words
- BitNet Style Quantisation: Storage vs Execution
- Face Based Attention: Gated Feature Mixing
- Hydraform: Adaptive Attention Head Mutation
- Judge Head Attention: Contextual Head Gating
- Transformer Activation Maps and Token Trajectories
- Transformer Memory: Causal Retrieval and Adaptation
- Self Healing Tokens: Selective Denoising Experiments
- Sparse Autoencoders: Reconstruction vs Feature Recovery
- Adaptive Neural Architectures: Experiments and Questions
- Temporal Attention for Irregular Observations
- SVD Compression: Storage and Functional Error
Supporting implementations accompany the articles. Each project README lists its entry points, dependencies and commands. The reproduction guide uses uv and pinned Python 3.12 dependencies to run the studies and their checks; browser tools identify their HTML entry point.