₦airaland Forum

Welcome, Guest: Register • Login • With Google • Trending • Recent • New

Stats: 3,344,118 members, 8,510,228 topics. Date: Friday, 30 October 2026 at 05:33 AM

Toggle theme

Taomole's Posts

Nairaland Forum › Taomole's Profile › Taomole's Posts

1 (of 1 pages)

Programming › Re: How I Optimized A Local Ai Vector Model To Run On A Cheap 8gb Vps (no Gpu) by taomole(op): 6:04pm On Jun 24
I'm curious, for those of you building e-commerce sites for clients in Nigeria, what framework do you typically use? WooCommerce, Shopify, or custom builds? I'm hoping to optimize our checkout integration next to make cart-merging seamless for local platforms.
Programming › How I Optimized A Local Ai Vector Model To Run On A Cheap 8gb Vps (no Gpu) by taomole(op): 3:08pm On Jun 24
Hello guys,

I wanted to share a project I just launched publicly today. It’s called AnonID, and it’s a lightweight, self-hosted alternative to FingerprintJS.

I built this because Safari's ITP and privacy updates are aggressively breaking standard cookies, making it hard to track returning anonymous users (for things like restoring abandoned shopping carts or personalizing sites pre-login). Commercial tools that solve this cost upwards of $99/month, which is unrealistic for most bootstrapped local projects and developers facing international card limits.

So, I decided to build a stateful, secure identity stitching engine that can run comfortably on a cheap, single 8GB Hetzner VPS.

Here is the technical stack I used to build it:

1. The Backend: Node.js & Express.
2. The Database: SQLite (using better-sqlite3). To prevent SQLite write deadlocks (SQLITE_BUSY) during concurrent traffic spikes, I configured it in WAL (Write-Ahead Logging) mode and set an active busy_timeout.
3. Concurrency & Scaling: I used Redis for atomic rate-limiting and distributed SET NX locks to handle incoming traffic safely.
4. The AI Vector Engine: I embedded an open-source ONNX model (all-MiniLM-L6-v2) to generate 384-dimensional mathematical vectors of page context for cross-device semantic matching.
5. Event Loop Protection: Because running CPU-bound WebAssembly math directly in Express would block the Node.js event loop and destroy our sub-20ms fast path, I offloaded the ONNX model completely to a persistent node:worker_threads background queue.
6. The Coffee Shop Collision Fix: Blindly merging users on the same Wi-Fi causes major data leaks (merging carts of two strangers). To solve this, I built a stateful soft-association algorithm that requires at least 3 distinct overlapping semantic sessions in a 15-minute sliding window before executing a canonical database merge.

The entire core engine is 100% open-source under the MIT license:
GitHub: https://github.com/NaijaB0T/anonid

And I set up an interactive landing page where you can run the console script and watch your live browser hardware JSON footprint resolve across cookie clears:
Website: https://anonid.pro

(Since my global payment gateway is currently pending verification, I’ve set up an instant Free Beta Key form directly on the landing page for launch day so you can test it out.)

I would highly appreciate your honest feedback on the database locking architecture, the worker thread implementation, or the privacy model!

Let me know what you think.

1 (of 1 pages)