Quartz 5

backend

4 min read

Systems / Backend Depth

These are the ones that actually make you dangerous as a backend engineer.

1. Build a mini reverse proxy (nginx-lite)

  • Features:

    • HTTP/1.1 parsing (raw, no frameworks)
    • Reverse proxy to upstreams
    • Load balancing (round robin, least connections)
    • Basic caching layer

  • Bonus:

    • TLS termination later
    • Rate limiting

👉 This aligns perfectly with your volk server — you can evolve it into this.

2. Write your own job queue (like Sidekiq / Celery)

  • Components:

    • Worker pool
    • Persistent queue (Redis or your own disk format)
    • Retry + backoff
    • Delayed jobs

  • Stretch:

    • Distributed workers over TCP
    • Priority queues

👉 This teaches real production backend patterns.

3. Build a simple database

Not a toy — something real enough to learn internals.

  • Start with:

    • Key-value store (append-only log)
    • In-memory index

  • Then:

    • Compaction
    • WAL
    • B+ tree or LSM tree

👉 This will change how you think about data forever.

4. Implement a file sync system (Syncthing-lite)

  • Watch filesystem changes
  • Send diffs over network
  • Conflict resolution
  • Resume transfers

👉 You’ve already had pain with Syncthing — perfect motivation.

5. Write a TCP-based chat server (no HTTP)

  • Raw sockets
  • Custom protocol
  • Multiple clients
  • Message broadcasting

Stretch:

  • Add encryption
  • Presence + rooms

👉 This builds real networking intuition.

🌐 Real-world Backend Apps (portfolio-ready)

6. Auth service (production-grade)

  • JWT + refresh tokens
  • OAuth (Google/GitHub)
  • Email verification
  • Rate limiting
  • Audit logs

👉 You literally have this on your Whisp backlog.

7. Analytics ingestion service

  • Accept events via HTTP
  • Batch + store efficiently
  • Query API

Stretch:

  • Use columnar storage
  • Add real-time streaming (Kafka-lite)

8. Multi-tenant SaaS backend

  • Organisations + users
  • Role-based access control
  • Rate limits per tenant
  • Billing hooks (mock Stripe)

👉 This is what companies actually build.

9. Webhook delivery system

  • Accept webhook registrations
  • Retry logic (exponential backoff)
  • Signing (HMAC)
  • Dead letter queue

10. Search engine (mini Elasticsearch)

  • Index documents
  • Inverted index
  • Ranking (TF-IDF)

🤖 AI-related backend projects (not just “call OpenAI API”)

11. LLM gateway (very relevant to you)

  • Single API to:

    • route to local models (llama.cpp)
    • fallback to cloud (OpenAI, etc.)

  • Features:

    • rate limiting
    • caching responses
    • prompt logging
    • cost tracking

👉 This fits your Hetzner + subdomain idea perfectly (ai.awais.me).

12. Vector database (mini Pinecone)

  • Store embeddings
  • Similarity search (cosine)
  • API layer

Stretch:

  • HNSW index
  • Disk persistence

13. RAG backend

  • Upload documents
  • Chunk + embed
  • Query with context retrieval

👉 Make it your own knowledge base replacement.

14. Code assistant backend

  • Input: codebase

  • Features:

    • semantic search
    • “explain this file”
    • “find bugs”

Stretch:

  • integrate with Neovim

15. AI task worker system

  • Queue tasks like:

    • summarisation
    • embeddings
    • code generation

  • Workers process using local LLMs

👉 Combine this with your job queue project = 🔥

⚙️ Infra / DevOps-focused backend

16. Service orchestrator (mini systemd / docker-lite)

  • Start/stop processes
  • Restart policies
  • Logging

17. Distributed rate limiter

  • Token bucket
  • Shared state across nodes (Redis or custom)

18. Config-driven deployment tool

  • YAML → deploy services
  • SSH + run commands
  • Health checks

🧩 How to pick (important)

Don’t try to do everything. Pick based on your goal:

  • Want deep systems knowledge → → database / proxy / TCP server

  • Want job-ready backend skills → → auth service / SaaS backend / webhooks

  • Want to align with your current stack + interests → → LLM gateway + job queue combo

🔥 My blunt recommendation for you

Given everything you’ve been doing:

Start this sequence:

  1. Job queue (core infra)
  2. LLM gateway (uses the queue)
  3. Extend your volk server into reverse proxy

That combo is:

  • practical
  • impressive
  • deeply technical
  • reusable in your own infra

If you want, I can map out a step-by-step build plan (with architecture + folder structure) for one of these so you can actually execute instead of just collecting ideas.

Graph View