Skip to content
Back to showcase

Inference Server

A low latency inference service for hosting AI models, built to keep the hot path fast and strictly typed.

Role
Solo, systems
Timeline
2025
Stack
Rust · gRPC
Status
Shipped
Inference Server

$ cargo run --release

Compiling gateway v0.1.0

Finished release target(s) in 8.42s

gRPC server listening on :50051

Fig. 01 · InterfaceRust · gRPC · Tokio

The problem

Model serving tends to accumulate glue: a slow layer here, an untyped payload there, and the hot path pays for all of it on every single request.

What it does

Fronts the models with a typed gRPC service, so the contract is checked at the edge and the request path stays as short as it can be.

Boring, typed, and fast beats clever every time.

PreviousOmni Turn DetectorCase study 02 / 03
Next project

Document Portal

Upload, parse, and find anything in seconds.

  • Next.js
  • Python
  • pgvector
  • OCR