Resource Hub

Resource Hub is a centralized platform for storing, organizing, and sharing learning resources — links, bookmarks, ordered collections, and documents.

The backend is built with Express 5, TypeScript, PostgreSQL, and MinIO. The React frontend handles resource management, public discovery, and collection profiles. Authentication is delegated to my shared Auth Service.

1. Auth Service Delegation

Instead of maintaining authentication logic inside Resource Hub, I integrated my shared Auth Service.

The backend expects an HTTP-only cookie containing a JWT signed by Auth Service. Tokens are verified locally using RS256 and JWKS — the same pattern used across my other personal products.

For a deeper explanation of the auth architecture, see the Auth Service case study.

2. Presigned URL Upload Architecture

Used presigned URLs for document uploads to MinIO.

Why?

Instead of:

Client → Server → Object Storage

Used:

Client → Object Storage

Reasoning

The server should not become responsible for transferring large files.

Benefits

• Reduced server load.
• Eliminated an unnecessary network hop.
• Improved scalability for file uploads.

The server still creates a PENDING database row and returns the presigned POST policy — it orchestrates the upload without proxying the file bytes.

3. Webhook-Driven Document State

Document uploads follow an async confirmation flow:

1. Client calls POST /api/document/upload-url → backend creates a PENDING row and returns a presigned POST policy.
2. Client uploads directly to MinIO.
3. MinIO fires a webhook to POST /api/document/webhook → backend marks the document SUCCESS (or handles failures).
4. Client requests a short-lived presigned GET URL to view the file.

Why?

A client-reported "upload succeeded" is not enough. The backend should only mark a document as available after object storage confirms the file exists.

Benefit

The document lifecycle reflects actual storage state rather than optimistic client assumptions.

4. PostgreSQL + Drizzle ORM

Migrated the data layer to PostgreSQL with Drizzle ORM.

Why?

I wanted a typed schema, SQL migrations, and a relational model that could express collections, collection items, bookmarks, and documents cleanly.

Benefit

Foreign keys, unique constraints (e.g. per-user resource deduplication, collection idempotency keys), and structured queries replaced document-store patterns that were harder to enforce at the application layer.

5. Idempotency Keys on Collection Mutations

Collection create and item mutations accept idempotency keys stored as unique constraints in PostgreSQL.

Why?

Adding items to a collection or creating a collection should survive network retries and double-clicks without duplicating rows.

Benefit

Safe retry behavior on collection writes. For a deeper treatment of idempotency mechanics in another project, see the Vaultex case study.

6. pg-boss Background Jobs

Used pg-boss (PostgreSQL-backed job queue) for background work:

• Link metadata fetch — on-demand Open Graph / metadata retrieval for resources.
• Pending document cleanup — nightly cron to remove stale PENDING documents older than a configured threshold.

Why?

Both jobs fit naturally on the same PostgreSQL instance already running the application — no separate queue infrastructure for a personal tool.

Limitation

Job throughput is bounded by a single PostgreSQL deployment. Acceptable at my current scale.

7. guestAuth vs optionalAuth

Two auth middleware variants serve different route needs.

• optionalAuth — requires a valid token (or dev bypass); used for most authenticated routes.
• guestAuth — allows anonymous access; attaches user context when a valid token is present and silently ignores invalid/expired tokens.

Why?

Public collection reads (/u/:username/:slug) should work without forcing login, while still personalizing the experience when a user is signed in.

Tradeoff

Simpler public sharing, but guest routes need careful handling so invalid tokens never block anonymous reads.

8. Low-Friction Create, Enrich-on-Edit

Resource creation accepts only name and link. Description, tags, and visibility are set on the edit page.

Why?

Requiring every field at create time adds friction and reduces the chance a user saves a link in the moment.

Benefit

Capture first, organize later. New resources are always created as private; enrichment happens when the user has time.

1. Subscription Model

Integrate a subscription model so users can upgrade to a paid tier on the platform when they need more than the free tier offers.