DEV Community: LOI CHIANG HAO

What's the first thing that breaks when a RAG system leaves the notebook?

LOI CHIANG HAO — Tue, 09 Jun 2026 07:12:11 +0000

I've been studying and building RAG systems for a while, and I've noticed a pattern.

Most tutorials stop once the demo works.
But production introduces a completely different set of problems.
In my experience, the first issues usually aren't model-related.
They're system-related.

Examples:

Retrieval returns technically relevant but practically useless context
Costs grow much faster than expected
Evaluation becomes difficult and inconsistent
Model updates introduce subtle behavior changes
Latency becomes a real constraint

The more I worked on RAG systems, the more I realized that many failures happen between components rather than inside the model itself.

That shifted how I think about AI engineering.

Instead of asking:
"Can this work?"

I started asking:
"What happens when it fails?"

I've been organizing everything I've learned into an open-source project called AI Model Atlas, focused on AI system design rather than specific frameworks.

GitHub:

🗺️ AI Model Atlas

👉 Read This First: Why AI-Model-Atlas Exists?

Open-Source AI Learning Map — From Zero to RAG, Agents & Fine-Tuning

📖 Bilingual docs (EN/ZH) · 36 curriculum modules · 17 deep-dive chapters · Runnable RAG sandbox
🎯 A learning-focused architecture simulator — not a production framework, not a live model catalog.

📌 Reading guide

✅ Great for: learning AI concepts, RAG system design, and hands-on experimentation

⚠️ Model names & API prices may change — always verify against official docs

🚫 Not intended for: production deployment or real-time benchmarking

🧭 Start Here

I want to…	Go to
📚 Learn from scratch (step-by-step)	CURRICULUM.md — 36 modules, Phase 1→5
🧬 Understand the math & internals	DEEP_DIVES.md — 17 chapters
📐 See the system architecture	ARCHITECTURE.md
▶️ Run the demo in 5 min	Quick Start ↓
🗺️ Pick a learning track	Getting Started Guide

📦 What's Inside

Content	Count

…

View on GitHub

For people who have deployed RAG systems:
What was the first production issue that surprised you?

I built a space where strangers leave messages for their future selves

LOI CHIANG HAO — Mon, 08 Jun 2026 08:30:30 +0000

Last night at around 3 AM, I wasn’t trying to build anything meaningful.

I was just testing a small experiment I had been working on.

A quiet digital space where people can leave messages anonymously.

No accounts.
No followers.
No feeds.

Just fragments of thoughts drifting in time.

Then I clicked on a random entry.

One of the messages said:

“Hope that one day I will learn to make peace with myself.”

I don’t know who wrote it, or when, or what they were going through.

But I stayed on that screen longer than I expected.

It felt different from how the internet usually feels now.

Everything today is optimized for:

attention
engagement
velocity

But this felt like the opposite.

Slow. Anonymous. Human.

So I expanded the idea into a simple system:

People can leave “time capsules” for their future selves.

Each message is locked for 365 days.

After that, there is a 30-day window where the author can choose:

publish it
or destroy it

If no action is taken, it automatically becomes public on day 396.

That’s where the name comes from.

No algorithm decides what matters.

Just time + uncertainty.

Sometimes you leave something into the void.
Sometimes the void eventually answers back.

It’s still very early, but I’m letting it exist.

If you want to try it:

👉

Day396 · Time Capsule · 时间胶囊

Cast your anonymous time capsule into the future. 向未来投递匿名时间胶囊。

day396.vercel.app

👉

I Built an Anonymous Digital Time Capsule Because I Miss the Old Internet

LOI CHIANG HAO — Mon, 08 Jun 2026 05:04:38 +0000

A few months ago, I started building a small experiment called Day 396.

Not another social network.

Not another AI tool.

Not another productivity app.

A digital time capsule.

The idea was simple:

What if people could leave messages for the future without profiles, followers, likes, or algorithms?

Just words.

Waiting.

Most of today's internet is optimized for immediacy.

Everything wants your attention right now.

Notifications.

Engagement metrics.

Infinite feeds.

Day 396 tries to do the opposite.

It encourages people to slow down.

Write something meaningful.

Seal it away.

And trust that time will do the rest.

Current Architecture

At the moment, the project is built with:

React (Vite)
Supabase PostgreSQL
Row-Level Security (RLS)
Anonymous write-only submissions
Dockerized reverse proxy infrastructure
Vercel deployment

One of the main goals was to make anonymous submissions secure while keeping operational complexity low.

What I'm Exploring Next

Some features are still ideas rather than shipped functionality.

I'm currently exploring:

Anonymous resonance interactions (🔥 💧 📡)
Better discovery mechanics for public capsules
Feature request workflows
Cross-language capsule relationships for curated seed content
Additional anti-spam protections

These are still being designed and may change significantly before implementation.

The Hardest Part

Interestingly, the hardest challenge hasn't been technical.

It's deciding what not to build.

Most platforms naturally evolve toward:

Trending content
Rankings
Recommendation algorithms
Social identity

I'm trying very hard not to go down that path.

The goal of Day 396 isn't consumption.

It's discovery.

I want every capsule to have a chance of being found.

If you're curious, you can try it here:

Day 396

Global:

Day396 · Time Capsule · 时间胶囊

Cast your anonymous time capsule into the future. 向未来投递匿名时间胶囊。

day396.vercel.app

China:

What Most RAG Tutorials Don't Teach You

LOI CHIANG HAO — Sat, 06 Jun 2026 09:26:55 +0000

Most RAG tutorials stop at something like:

Vector Search → LLM → Done

And for learning the basics, that's completely fine.

The problem is that once systems become larger, several additional layers start to matter:

Query routing
Hybrid retrieval
Semantic caching
Evaluation and feedback loops
Failure handling and fallback logic

These topics are often mentioned briefly, if at all, in beginner tutorials, yet they can have a significant impact on cost, reliability, and user experience.

Why I Built AI Model Atlas

I wanted a way to study and visualize these architectural patterns without turning them into another framework.

So I built AI Model Atlas, a learning-focused repository that explores concepts such as routing, hybrid retrieval, caching, evaluation, and execution control through guided modules and runnable examples.

GitHub:
https://github.com/Hao610/AI-Model-Atlas

The goal isn't deployment.

The goal is understanding how production-style AI systems are structured and why the simple tutorial version often isn't enough.

It is designed as:

A learning-focused architecture simulator
A guided curriculum (36 modules)
A reference architecture for RAG system design
A conceptual bridge between tutorial systems and production thinking

Discussion

I'm curious:

What layers have you found most important when moving a RAG system from demo to production?

Routing?
Evaluation?
Caching?
Observability?
Something else?

DEV Community: LOI CHIANG HAO

What's the first thing that breaks when a RAG system leaves the notebook?

Hao610 / AI-Model-Atlas

Bilingual open-source AI learning map: RAG, agents, fine-tuning & system design (EN/ZH) | 中英双语开源 AI 学习路线图：RAG、智能体、微调及系统设计

🗺️ AI Model Atlas

Open-Source AI Learning Map — From Zero to RAG, Agents & Fine-Tuning

🧭 Start Here

📦 What's Inside

I built a space where strangers leave messages for their future selves

Day396 · Time Capsule · 时间胶囊

I Built an Anonymous Digital Time Capsule Because I Miss the Old Internet

Day396 · Time Capsule · 时间胶囊

What Most RAG Tutorials Don't Teach You

Why I Built AI Model Atlas

Discussion