An agent that drives the whole ML lifecycle through a 20+ tool MCP registry behind human approval gates, running every model-generated cell inside a hardened Docker sandbox with no network, a read-only rootfs, and a non-root user. The preprocessing path is marked a scaffold, not claimed as shipped.

A multi-stage NLP smart-input pipeline (chrono-node, compromise, and a custom priority parser) that resolves overlapping parser claims by priority first, then per-tag confidence. The serverless path dropped Prisma for raw pg, while a parallel Express + Prisma stack still ships alongside it.

An MLP whose hot paths are hand-vectorized with arch-dispatched AVX-512, AVX2, and NEON kernels over 64-byte-aligned, blocked-tiled matrices, no BLAS or PyTorch pulled in. OpenMP wins on big matrices and regresses small ones ~7x, so it is gated by size, not waved off.

A disk-backed relational engine built end to end: SQL parser, cost-based optimizer, Volcano executors, MVCC with two-phase locking, and a WAL with ARIES recovery. It beats SQLite on batch inserts and loses to it on point selects by 2.0 to 2.6x, reported, not hidden.

A fine-tuned DeepSeek-Math 7B paired with FAISS retrieval over NaturalProofs and a SymPy tool registry, driven by an autonomous ReAct loop. It reports the full ablation ladder, 41.2% prompted up to 77.6%, and flags those figures as documented sample data, not a recorded GPU run.

28 C++23 containers that self-host on each other the way a real standard library does: map on a red-black tree, lru-cache on list plus hash map. flat_map runs 0.378x of std::map, and the tree map sits within 12 to 18% of std, written down as a loss instead of buried.

A C++23 expression language compiled to WebAssembly: tokenizer, recursive-descent parser, tree-walking evaluator, user functions, and recursion. It is the engine parsing the headline at the top of this page. It uses double everywhere with exact == checks, simple and stated outright.

One generic maze core, GenericMaze over a C++20 GraphCell concept, where each algorithm is a single templated method that drives both the headless solver and the live terminal animation. The callback copies the whole frontier each step, an O(n) cost taken on purpose for legibility.

A symbolic differentiation engine in pure Python with exact rational arithmetic and polynomial GCD, so derivatives stay exact instead of drifting to floats. Rational layer and GCD are hand-rolled rather than imported from fractions, the whole point and the cost of the project.