How AI Helped Me Build a Lightweight Pseudo‑Full‑Stack WeChat Mini‑Program

Preface

Recently, with the help of AI, the author became a "pseudo‑full‑stack" developer, building a WeChat mini‑program despite limited prior full‑stack experience.

1. Keep features simple

Because full‑stack work requires handling UI, front‑end, back‑end, and workflow simultaneously, the author stresses that simpler features are easier to implement and modify.

2. Multiple mini‑program versions coexist

After release, a minimum version set in the WeChat backend remains active while updates roll out gradually, leading to several versions running concurrently. This requires extensible data structures.

Data structure : design a main table and related tables to allow field extensions for multi‑version support.

API versioning : version APIs; once the mini‑program upgrades successfully, old APIs can be safely deprecated.

URL version prefix : follow a Kubernetes‑style pattern, e.g. /api/v1/xxx, adding /api/v2/ for new iterations while keeping old endpoints until they are no longer used.

3. Cloud database overhead

The project uses TiDB Cloud (Singapore region). Tests show SQL round‑trip latency near 200 ms, far higher than the tens‑of‑milliseconds typical of internal networks.

Optimization strategies include:

Window function merging : replace separate COUNT(*) and data queries with COUNT(*) OVER() AS total to retrieve total count and page data in a single SQL, reducing one network round‑trip.

Batch inserts : aggregate multiple rows into a single INSERT statement, dramatically cutting network overhead.

Push‑down computation : leverage TiDB Cloud’s support for window functions and other features to perform more calculations on the database side instead of pulling data to the application.

4. Mini‑program tech stack

The current stack uses uniapp together with the wot‑ui component library, which simplifies development for mini‑programs.

While uniapp was chosen for potential multi‑platform support, the author notes that a native WeChat implementation could be lighter, though rewriting the existing project is not feasible.

5. Stay patient

After deployment, the author only monitors essential security and error logs, otherwise lets the program run unattended to maintain a good mindset. User activity gradually appeared, providing a pleasant surprise.

6. Backend language choice

When using AI‑generated code, the author finds Rust and Java code generation efficient, whereas Python generation is problematic. Rust is preferred because:

AI language efficiency : AI produces high‑quality Rust and Java code, but struggles with Python.

Resource consumption : Rust runs with a smaller footprint than Java, which still consumes considerable resources even when lightweight.

Code style : avoiding excessive abstraction keeps the project simple and easier for AI to understand and maintain.

7. Targeted optimization

Previously, optimization focused only on the backend. In a personal full‑stack project, the author now considers the entire stack, recognizing that changes often require coordinated front‑end and back‑end updates, so the focus is on solidifying each side individually while providing fallback mechanisms.

8. Security cannot be ignored

Full control over host and application makes security unavoidable. The author’s security practices are divided into four layers:

Server layer : enable ufw firewall, open only necessary ports, and configure basic Nginx rate‑limiting to block malicious scans.

Daily operations : regularly review logs and tighten rules when suspicious access is detected.

Application layer : add authorization checks to ensure users can only manipulate their own data.

AI assistance : let AI implement most security measures and periodically scan the project for security recommendations.