How far down the stack I've been.
Most of my career has been near the surface: interfaces, APIs and the glue between them. I'm going lower, one layer at a time, and I'm keeping the score in public. Each layer below says what exists and what doesn't.
- Shipped
- In progress
- Planned
Layers, from the interface down to the kernel
- 01 / InterfaceShipped
Products people use
What existsZamir (web and mobile), CoinSafe's app and wallet onboarding, and client sites, all in production.
What's missingNothing blocking. This is where most of my four years went.
- 02 / ServicesShipped
APIs, auth and data
What existsNode API routes, Postgres through Supabase, Keycloak authentication and email, behind real products.
What's missingA Rust service in production. Tokio and concurrency are still more theory than practice for me.
- 03 / ProtocolsIn progress
ZK proof systems in Rust
What existsGKR, KZG commitments and Sumcheck, with polynomial arithmetic, Fiat–Shamir, tests and a Criterion benchmark harness.
What's missingPublished benchmark numbers with their method, and a write-up of each protocol's design decisions.
- 04 / MemoryIn progress
Allocators
What existsA bump allocator, implemented.
What's missingThe linked-list (free-list) allocator, tests, and benchmarks against the system allocator. Not finished, so not presented as finished.
- 05 / KernelIn progress
An operating system in Rust
What existsKernel fundamentals and memory management, following the blog_os approach, plus data-structure practice in rust-learn.
What's missingThe no_std implementation isn't complete yet.
The next three.
I work in milestones at my own pace, not on a calendar. A milestone is finished when its gate is met and the artifacts are public, and not before.
- Milestone 01
Production systems and memory
- Bump allocator (Done)
- Free-list allocator with splitting and coalescing (In progress)
- Benchmarks against the system allocator (Not started)
- Async networking server with graceful shutdown (Not started)
- Rust backend with Axum, SQLx and Postgres, deployed (Not started)
GateAllocator and server documented in public, with benchmarks and flamegraphs. - Milestone 02
Distributed systems and storage
- Raft leader election and log replication (Not started)
- Write-ahead log and snapshots (Not started)
- Failure tests under simulated partitions (Not started)
GateA cluster that survives leader loss and partitions, with the failure analysis published. - Milestone 03
AI infrastructure
- Per-generation cost and failure tracking in Zamir (Not started)
- Inference gateway with batching and streaming (Not started)
- Load tests reporting p95 and p99 (Not started)
GateA gateway with published tail-latency benchmarks, driven by real usage patterns.
What “done” means.
- A design record
- Why this design, and what I rejected.
- Reproducible benchmarks
- Latency percentiles and throughput, with the harness in the repo.
- Failure analysis
- What happens under partitions, drops or memory pressure.
- Profiles
- Flamegraphs showing where the time and cache misses go.
- Diagrams
- Dataflow and state transitions you can follow without the code.