Marcus Tancredi
Senior engineer · infra
and developer tooling.
Berlin-based, staff-level engineer with ten years of production experience. I work on infrastructure, developer platforms, and the tooling layer that quietly decides how fast a product team actually ships. Previously Shopify, Supabase, and a Series-B deeltech startup you'd recognise.
Five projects, sharp edges.
Remote dev-container platform
Led the infra for Arc Foundry — Firecracker microVMs, Nix-based image pipeline, 140ms p99 cold-start across EU + US regions. Shipped from zero users to 1,900 paying teams in 14 months. Now serving 4.1M container-hours/month.
RustFirecrackerNixNATSClickHouseBuild-graph compiler for the storefront API
Replaced a 40-minute monorepo build with a content-addressed graph compiler. Result: 94-second p50 full build, 2.1-second cached build, <1-second incremental. Open-sourced the query planner; still in production on Shopify's storefront API today.
GoBazelBuck2gRPCpgvector hot-path optimisation
Three-month engagement: benchmarked and optimised pgvector's HNSW index build on large-document corpora. Cut p95 insert latency from 340ms to 18ms on a representative 12M-row workload. Upstreamed four patches to pgvector.
PostgrespgvectorCRustObservability pipeline rewrite
Rewrote our log/metric ingestion layer from Kafka+ES to a column-oriented ClickHouse + Vector pipeline. Infra cost down 67%; p95 query latency on 18-month data under 420ms. Acquired by Datadog in Q4 2022.
VectorClickHouseKafkaGosparse-cache · a Rust library
Lock-free, sharded LRU with per-shard compaction. 11k GitHub stars. Used in production by two companies I'm not allowed to name and a handful of databases I am: EdgeDB, TurboPG, and a rather large key-value store.
RusttokiocriterionWhat I actually use.
LANGUAGES
- Rust
- Go
- TypeScript
- Python (data / scripts)
- SQL (Postgres flavour)
INFRA & PLATFORM
- Kubernetes · Nomad
- Firecracker · containerd
- Terraform · Pulumi
- AWS · Fly · Cloudflare
DATA
- Postgres · pgvector
- ClickHouse · DuckDB
- Kafka · NATS · Redpanda
- Iceberg · Parquet
OBSERVABILITY
- OpenTelemetry
- Prometheus · VictoriaMetrics
- Grafana · Vector
- Honeycomb for tracing
Ten years, quietly.
I started writing code professionally in 2016 at a logistics startup in Milan, moved to Shopify's Montréal office in 2020, and settled in Berlin in 2023 after a year of contract work across Europe. I have a master's in distributed systems from Politecnico di Milano and spent a year of it at TU Berlin on exchange.
I work mostly solo or as an embedded staff engineer inside small teams (3–8 engineers). I've led two small teams of my own and found out I'd rather not. I'm best early: 0-to-1 infra, platform foundations, the first serious performance work. Comfortable in production at any size; less interested in pure green-field R&D.
Outside of work I ride an ageing BMW motorcycle, cook badly, and maintain sparse-cache.
Recent essays.
What I wish I'd known before writing a database cache
Eighteen months of mistakes with sparse-cache — ordered by how much pain they caused.
Read →FEB 02, 2026Build graphs don't have to be slow
What we learned replacing 40 minutes of Bazel with 94 seconds of custom graph compiler.
Read →NOV 11, 2025Four things Firecracker won't do for you
Cold starts are easy. Networking, rootfs lifecycle, and bin-packing are the real work.
Read →Let's build something
quiet and fast.
I'm open for 1–2 engagements in summer 2026. Typical shape: 3–6 months, embedded with a small team, focused on platform / infra / performance work. I do not do one-off MVPs, CTO-for-hire, or anything where I'd be the only engineer.
hi@marcus.devgithub.com/marcus-t@marcus.11