Marcelo Lazaroni

lazaroni.marcelo@gmail.com · +351 916 255 309 · github.com/lazamar · linkedin.com/in/lazaroni · Porto, Portugal

I work on compilers and database internals. I was one of the two people maintaining the core of Glean, Meta’s graph database of facts about code; its query language compiler in Haskell and its runtime in C++. I have since led engineering at an event-sourcing infrastructure startup.

I like problems of correctness at scale, where it takes good abstractions to make systems both fast and safe.

Experience

Ambar — Principal Engineer ➔ VP of Product

2023 – present · Haskell, TypeScript, AWS

Founding engineer. Led product engineering and direction as the company grew from 4 to 19 people, managing 8 developers across two areas.

• Set company and product direction with the CEO, deciding what to build next and which clients to take on.
• Hired the engineering team. Ten senior engineers from 50+ interviews. Took them from no event-sourcing experience to shipping independently by building the first product with them — an AI assistant for estate agencies, live for 15+ agencies in three months — reviewing 305 of their pull requests.
• Designed the type-driven event-sourcing framework behind every client application we built. The compiler enforces domain constraints, so new hires and AI assistants write more of the code with fewer errors. Eight applications shipped on it within a year.
• Let prospects run the entire product on a laptop. A single static binary standing in for the whole platform: partitioned, durable, ordered queues and connectors for three databases. 10,000 writes in 29 ms; 10,000 reads in 1 ms.
• Built a filtering DSL to support our largest client’s volume. Filters evaluate against encoded data, never deserialising it. 60,000+ messages per second per core. Typechecked against schema files and deliberately not Turing-complete.
• Enabled language-agnostic queue consumption at full speed, no SDK needed, by building our push protocol’s backpressure mechanism, which infers client load from response times and sizes a concurrency window to match. Model-checked and regression-tested in CI.
• Taught event sourcing to 300+ engineers in weekly five-hour courses run to build demand.

Meta — Software Engineer (IC5)

2021 – 2023 · Haskell, C++ · Code Search Infrastructure

Built core features of Glean, a graph database of facts about code, and of the typed Datalog-like language used to query it. Glean sits behind IDE tooling, code navigation and static analysis company-wide.

• Designed the evaluation strategy for recursive queries, Glean’s most requested feature, to enable transitive dependency and points-to analysis over all of Meta’s code. Bottom-up semi-naive evaluation with a modified Magic Sets transformation. Grounded in the Datalog and tabled-logic literature.
• Enabled large-scale dead code detection by adding negation to Glean’s query language. Reference-existence analysis became a database query, leading directly to the removal of 26,000 lines of dead Python and C++. It also underpins company-wide symbol search.
• Took schema migrations from over two months to under a week with automatic backwards- and forwards-compatible schema changes. Migrations had needed separate pipelines and coordination across 17 clients; the feature cleared a backlog of blocked changes.
• Made incremental indexing viable. I sped up a key indexing step for Hack, Meta’s PHP dialect, from over 80 minutes to under 3, and total creation from 120 minutes to 40; index freshness went from 4–5 hours to 1–2.
• Resolved cluster overload without new hardware with a query-compilation optimisation that prunes unproductive branches: P95 compilation fell from 27 ms to 2.5 ms, and the heaviest client’s load dropped sevenfold.
• Unblocked the company’s response to the Log4j vulnerability. The build system ran out of memory enumerating affected targets; I wrote the Glean script that found all of them, with the owning team for each.
• Eliminated a recurring source of production instability by implementing request prioritisation in the open-source Haskell Thrift library — server restarts from missed health checks went from 100 a week to zero, everywhere the library is used.

Standard Chartered Bank — Quantitative Developer

2019 – 2021 · Haskell · Strats, Financial Markets

Owned the interest-rate swaps section of a million-line derivative pricing service, and the relationship with the traders it served.

• Implemented real-time trades pricing, going from on-demand swap pricing to streaming prices multiple times per second.
• Made upstream data validation fast enough to run continuously by removing a 2ⁿ factor from its time complexity — found by auditing the code and spotting a property of the calculation that made it tractable.

Pushfor — Lead Frontend Engineer

2018 – 2019 · TypeScript, Elm

Promoted to lead four months after joining, after rebuilding the most error-prone part of the product.

• Reduced new bugs tenfold and ended debugging sessions with clients, by making the case for restructuring the frontend and leading three developers through a staged migration.

Code Dock — Director

2017 – present · TypeScript, Elm

A one-person company, founded for the project below; now runs paid internships for new programmers.

• Won, negotiated and delivered a CRM platform for a financial consultancy managing over £5bn in client assets, later expanded to cover expenses, project leads and time tracking.

Selected projects

• Nix Package Versions — find every version ever available for any Nix package. Haskell.
• haskell-docs-cli — browse Hackage from the terminal.
• lazamar.github.io — writing on compilers, performance and functional programming.

Education

BSc Computer Science, First Class Honours — University of West London, 2013–2016. Awarded scholarships for academic performance in 2013 and 2014.