Deep-Dives

A comprehensive local RAG stack (ChromaDB + Ollama) designed for strictly private, reproducible retrieval and LLM inference; heavily focusing on hybrid retrieval strategies and index versioning.

A highly optimized Sudoku engine engineered in Rust, featuring advanced human-like techniques, multi-platform support (Python, WASM), and microsecond-level performance.

A from-scratch, Rust-implemented compiler designed for an expanding C subset (user-defined structs, multidimensional arrays, pointer semantics, rich control flow, standard headers) that emits verified LLVM IR and links native executables via clang.

A from-scratch Git implementation in Rust; exploring content-addressable storage, plumbing/porcelain layering, high-performance LRU caching, .gritignore filtering, and Criterion benchmark suites.

A hands-on ML workshop that grew from a CLI script runner into a full-stack application: a FastAPI async backend dispatches training jobs to a thread pool, intercepts matplotlib output in-memory, and streams live telemetry to a Next.js dashboard via Server-Sent Events, while still supporting headless CLI experimentation with uv.

The dock's HDMI output went black on one Mac. Testing pointed to an interaction between macOS 26, the monitor's HDMI Deep Color setting, and the dock's HDMI converter; direct DisplayPort provided a workaround.

A robust Java-based photo organization and migration engine featuring content-hash deduplication, a metadata-driven folder structure DSL with fallback chains, XMP sidecar discovery, parallel transfer pipelines, and multi-backend cloud targets.

A highly effective, local-first email auditing and automated cleanup tool designed to definitively identify noisy senders and deliver actionable, strictly privacy-preserving recommendations.

A comparative study of Apple-native frameworks (Vision, AVFoundation) against cross-platform C++/Python (OpenCV, pybind11) for video feature extraction.

An empirical systems exploration of local LLM inference on unified memory architectures: profiling compute-bound prefill vs. memory-bound decode phases, analyzing KV cache retention during cancellation, measuring the 64K vs. 128K context cliff on Apple Silicon, and comparing client payload strategies.

A deep architectural comparison of data processing pipelines: evaluating Apache Spark's batch ETL model against Apache Beam's portable unified model and Apache Flink's native API for stateful processing.

A decade-spanning food discovery app migrated from Ruby Sinatra to Rails to Django; documenting the real costs of incremental framework migration, ecosystem drift, and the discipline of staying current across 10+ years of active maintenance.