Soul Archives — Developer Story

Inspiration

It began with a conversation late at night. A friend said to me, "I feel terrible today, but I don't know what that feeling is."

That moment struck me deeply. We live in an era obsessed with "emotional intelligence." Social media feeds are flooded with articles like "10 Types of Anxiety You Must Know." But the truth is, most people cannot accurately name what they are feeling. When asked "Are you okay?", the most honest answer is often simply: "I don't know — I just don't feel right."

Naming emotions is a skill that takes practice. But before anyone can practice, they first need something much simpler: a safe, low-barrier outlet to speak.

So I asked myself: what if an app never asked "What emotion are you feeling?" but only asked "What's on your mind?" What if it didn't demand users to become their own therapist, but instead — like an old-fashioned detective — approached those unspoken feelings with curiosity and respect, tracing the clues hidden in plain words?

That is how Soul Archives was born.

What I Learned

1. Technology serves humanity, not the other way around

This project taught me that the best technology does not demand precision from the user. It meets them where they are. A user who types "I feel like I'm trapped in a glass box" should receive the same care as someone who types "I am lonely." The AI's job is not to correct — it is to understand.

2. Designing for emotional depth requires restraint

Every screen, every word, every animation must earn its place. A single unnecessary button can break the delicate atmosphere of trust. I learned that in emotional design, what you remove is just as important as what you keep. The final homepage has almost nothing on it — and that emptiness is intentional. It says, "There is space for you here."

3. Conversational development is a new paradigm

Building with MeDo through natural language felt less like coding and more like directing. I could describe the feeling I wanted a feature to have — "make the loading screen feel like someone is patiently waiting with you" — and the platform would translate that into technical reality. This blurs the line between designer and developer in a way that fundamentally changes how products can be built.

How I Built It

The entire application was constructed through iterative conversation with MeDo.

• Initial scaffolding: I started with a detailed prompt describing the core loop — emotional input, AI watercolor generation, detective story creation, and an archive card. MeDo generated a working full-stack application in minutes.
• Emotional analysis layer: The biggest leap came when I realized users needed help identifying their emotions. I asked MeDo to insert an AI-powered listening and empathy step before any art was created. This one addition transformed the product from a novelty into something genuinely supportive.
• Parallel generation and speed optimization: I noticed the watercolor and story were being generated sequentially. Through conversation, I restructured the flow so they trigger simultaneously, and added streaming responses to eliminate perceived waiting time.
• Full English localization: Just before submission, I discovered the competition required English. One final prompt translated every user-facing string while preserving the app's gentle tone.
• Debugging through dialogue: When the preview broke, I simply described the issue to MeDo, and it diagnosed and fixed the problem — no console diving required.

Challenges I Faced

1. Time constraint and credit limit

With only two days and a limited number of free credits, every prompt had to count. I learned to combine multiple optimizations into single, dense prompts — merging speed fixes, feature additions, and copy changes into one instruction. This constraint forced a discipline that ultimately improved the product.

2. Preserving emotional tone during translation

Moving from Chinese to English was not just a linguistic challenge but an emotional one. Phrases that felt warm and intimate in Chinese could sound clinical in English. I had to carefully rewrite each string to ensure the app still felt like a person who genuinely cared, not a machine.

3. Making AI feel human

The biggest technical and design challenge was making the AI's responses feel personal rather than generic. The dynamic empathy feature — which generates a unique encouragement and a public figure quote matched to the user's specific emotion — was the solution. Getting that to work within MeDo's orchestration required precise prompt engineering, but the result was worth it: every user receives something written for their moment alone.

---

Built entirely on MeDo over 48 hours. The simplest tools, for the most human experiences.

Built With

• ai
• ai-text-generation-api
• css
• generation
• html
• image
• javascript
• medo
• node.js
• postgresql
• react
• supabase
• typescript