NEURAL-NOTES

This is monorepo of neural-notes app.

Structure

Write-back Cache as an Architectural Trade-off

axum-redis-cache A write-back cache layer was introduced in front of the database using Redis and evaluated within this product, a personal document web application.

In practice, this approach resulted in measurable performance improvements:

• Significantly reduced read and write latency
• Removed database contention from the synchronous request path
• Enabled fast responses without additional latency-hiding techniques

Limitations

The limitations of this approach are as follows:

• Application state becomes difficult to reason about if the Redis cache layer is unavailable or inconsistent
• Strong durability guarantees cannot be provided without additional recovery mechanisms
• Therefore, this model is unsuitable for domains where data must be durably preserved (e.g. document applications)

In summary, a write-back cache is effective for high-throughput, latency-sensitive workloads, but requires careful domain selection and should be applied only with careful consideration of the domain.

Quick Start for dev

1. install just & docker

2. in your shell

   git clone git@github.com:lyh4215/neural-notes.git
   cd neural-notes
   just env
   just dev

3. go to http://localhost:5173

Backend - axum

• use axum-redis-cache

Frontend - react

Embed api - FastAPI

• use embedding for related post recommanding system