Tam Nguyen Duc

Go toolchain that turns research papers into tagged Markdown in four languages, plus the reading app that serves them. OCR, LaTeX, figures, references, translation and an eighty-four rule audit.

A hundred landmark computer science papers as tagged Markdown in English, Vietnamese, Chinese and Japanese. Corpus only, never a PDF.

Cross compilation for rucc: the target model, the ABI descriptions, the libc stubs, the sysroots and the harnesses that test all of it

Real world C projects, built unpatched with rucc and graded against GCC 16. A ladder of six rungs from a single file to SQLite, with the project's own test suite as the oracle.

Ship the decoder with the data. Self-decoding datasets in Rust, with WebAssembly decoders that any engine can run.

The text of Tolkien Gateway rebuilt as Markdown, with full front matter and a source link on every page

Go toolchain that fetches Tolkien Gateway, converts it to Markdown, keeps it in sync with the live wiki, and publishes it as a website and an EPUB.

The toolchain that builds tamnd/arxiv: harvest arXiv metadata, extract papers into tagged Markdown with the mathematics intact, translate, connect, and publish as web, EPUB, TeX and PDF.

A corpus built from arXiv: metadata for every paper, licence-permitting papers extracted into tagged Markdown, translated into Vietnamese, Chinese and Japanese, and published as web, EPUB, TeX and PDF.

Transport, routing, queueing and record keeping for running long jobs across a mixed fleet of language models. Stdlib only.

The benchmark harness for rudb. ClickBench, TPC-H, TPC-DS, JOB and CEB against DuckDB, ClickHouse, Umbra and DataFusion, with the reporting rules that decide what a number is allowed to claim.

The DuckDB compatibility harness for rudb. Differential execution against a real DuckDB, weighted function and statement coverage, storage format round trips, and C ABI conformance.

An embedded analytical database written in Rust, compatible with DuckDB. Its own columnar format with global dictionaries and multi-column compression, and operators that run directly on the encoded data. The bar is 10x DuckDB on ClickBench at a tenth of the disk.

A systematic C corpus for rucc. Every program targets one named optimization and carries the answer computed in Rust, so a bug that gcc and rucc share still fails.

A cache server in Rust. Swiss-table shards, a per-shard slab arena, RESP2/RESP3. Apache-2.0.

The pandas conformance suite for firepanda. Every pandas callable, every parameter, run against pandas 3.0 and compared under written rules. No skip outcome.

Reproducible benchmarks for columnar storage formats, and a standing attempt to re-run the numbers the papers published.

Throughput, latency and CPU cycles for seven cache servers, measured with memtier_benchmark. A Rust port of tidwall/cache-benchmarks that adds yo, runs on hardware that is not a 32 core AWS box, and draws all 154 charts deterministically from committed data.

Core libraries for Mojo, modelled on the Go standard library. JSON, XML, regexp, compression, archives, images, crypto and TLS, sockets and DNS, database/sql, time zones, templates, concurrency and testing. One repository, many packages you can depend on independently.