AS

adsharma

Arun Sharma — systems engineer, graph database founder

LadybugDB maintainer ex-Facebook kernel ex-Google
48
Comments
~4 mo
Date range
8
Threads
~3
Avg depth
Graph databases
Memory allocators / kernel
Python transpilation
LLMs & AI systems
Columnar / embedded DBs
Open-source licensing

Arun Sharma is a senior systems engineer turned founder. He spent time at Facebook maintaining the Linux kernel tree (jemalloc/MADV patches, c. 2011–2013) and contributed to Google-wide profiling infrastructure. He's now building LadybugDB — an open-source embedded columnar graph database forked from KuzuDB, with a focus on Cypher queries over Parquet/DuckDB storage.

He's active on HN as an operator-founder: commenting with first-hand war stories, linking his own GitHub discussions and blog posts, and occasionally sparring with people who disagree with him about graph DBs or jemalloc patches.

"LLMs are themselves a graph database with probabilistic edge traversal. Some apps want it to be deterministic."
"Vector + keyword search gets you 70–80% on evals. Graph handles the last 20–30%."
"Transformers are GNNs which won the hardware lottery."
"Write static Python, transpile to Rust/PyO3. Use a GC'd intermediate language until you have millions of users."
"Rewriting a complex working piece of software in Rust is not trivial — an incremental path is better."
drops own GitHub links often cites Anthropic & Google research argues with ex-colleagues publicly thin-skinned about downvotes strong opinions, lightly hedged 10+ year idea incubation cycles

He publicly noted a "consistent pattern of downvoting" and linked an X post about it rather than adjusting his tone. He also acknowledged openly: "Guilty as charged of carrying ideas in my head for 10+ years and acting on them later."

LadybugDB — github.com/LadybugDB/ladybug — embedded graph DB (Cypher + Parquet/DuckDB)
pgembed — fork of pgserver (pglite) with vector + BM25 extensions
py2many / v-ast — Python AST → multi-language transpiler (V-lang backend)
mcp-server-ladybug — MCP server wrapping LadybugDB for GraphRAG
adsharma.github.io — personal blog (perf hints, DuckDB wikidata compression)
blog.ladybugmem.ai — longer-form LadybugDB writing
Grafeo – A fast, lean, embeddable graph database built in Rust (March 2026)

~20 comments in a single thread, positioning LadybugDB against Grafeo, defending graph DB legitimacy, debating CSR-on-disk vs WCOJ, and defending his choice not to rewrite in Rust. Classic founder mode: every reply is both a technical argument and a sales pitch.

generated from hn.algolia.com/api/v1 · adsharma · march 2026