selected projects

A non-exhaustive list. Things I work on now or recently. Almost everything below is on Codeberg with a GitHub mirror at github.com/gburd.

PostgreSQL extensions

pg_mentat

Datomic-compatible Datalog query engine running entirely inside PostgreSQL. Implements Datomic's data model — immutable facts (datoms), schema-first attributes, the full Datalog query compiler, the pull API, time travel, and ACID transactions — accessible from any PostgreSQL client. Built with pgrx. A continuation, in spirit, of Mozilla's abandoned mentat project.

pg_infer

A PostgreSQL extension that exposes transformer model knowledge — gate activations, feature labels, learned associations, embeddings — as SQL-queryable relations and a custom index access method. Inference becomes a planner-visible operator the planner can cost, schedule, and parallelize, not an external service the database calls. Status: experimental.

pg_tre

Native PostgreSQL 18+ index access method for approximate-regex matching. Three-tier filter funnel (range bloom → sparsemap-backed trigram postings → per-tuple bloom) with a TRE-library recheck. Genuine Levenshtein-distance regex with WAL coverage, VACUUM awareness, and REINDEX CONCURRENTLY support.

Database / query engine work

Ra — relational algebra rule system

A query optimization framework built on literate programming, equality saturation (using egg), and differential dataflow. 1,387 transformation rules across logical, hardware (GPU/FPGA/SIMD), distributed, multi-model, and physical categories — each as a documented .rra file with formal algebra, preconditions, cost model, and tests. Includes a learned cost model trained with online execution feedback.

DBSQL

A small SQL92 layer on top of Berkeley DB. Inspired by SQLite, diverged in 2003, mothballed for a decade, currently being revived. Latest: v0.4.0, January 2026.

bdb-rs

Berkeley DB in Rust. Early-stage port; concurrency control and the on-disk format are the interesting parts.

Libraries

sparsemap

A sparse, compressed bitmap library for C, optimized for workloads with long runs of consecutive set or unset bits. 16 KB of consecutive set bits in 8 bytes; worst case (random bits) is a raw bitmap plus 8 bytes of overhead. Used as the postings layer in pg_tre.

skiplist

A header-only, lock-free C skip-list (specifically a splay-list: skip-list with optional adaptive rebalancing). Fraser/Harris marked pointers, epoch-based reclamation. Implemented entirely in preprocessor macros — type-specific code generated at compile time.

Lime

A runtime-extensible LALR(1) parser generator. Like Yacc/Bison — but the generated parser can load and unload grammar extensions at runtime without recompilation. SIMD-accelerated tokenization (AVX2/NEON). Used for the regex-pattern parser in pg_tre.

Hegel

A universal property-based testing protocol. A Python (Hypothesis) server speaks to client libraries in Rust, Go, C, and TypeScript via Unix sockets. One PBT engine, many languages.

PostgreSQL community infrastructure

pg.ddx.io

A community-facing portal indexing 30 years of PostgreSQL development: pgsql-hackers and related list archives, the upstream/master git history, the build farm, the commitfest, plus the pre-PostgreSQL UC Berkeley archives at /m/legacy. Backed by an Agora MCP server exposing 50+ tools for git/email/code-intelligence to AI agents.

Older / fork lineage

mentat

Original Rust port / continuation of Mozilla's mentat embedded Datalog store. Now superseded by pg_mentat for most uses.

Pure language LLVM Orc port

Albert Gräf's term-rewriting language with an LLVM JIT backend. Updated to current LLVM Orc API.

pcompress

Maintenance / modernization of Moinak Ghosh's archiver. Variable- block dedupe, polynomial fingerprinting, MinHash similarity, adaptive compression by file type. v4.x build system + multi- architecture support.

Infrastructure / forks

Maki (fork)