Skip to main content

The behind-the-scenes story of how we chose our tech stack, what went wrong, and why we changed course

The Tech Stack We Chose (And Why We Switched Midway)

The Tech Stack We Chose (And Why We Switched Midway)

By Rexman
Published: 13/06/2025 – Behind the Scenes Series #2

🧠 The Master Plan

Choosing a tech stack was easy. Or so we thought. We had a Notion page comparing Postgres vs Mongo, Firebase vs Supabase, React vs Vue vs SvelteKit. We were acting like we were choosing our life partner — when really, we just needed a stack that wouldn’t break in 3 weeks.

We picked:

  • Frontend: React Native with Expo
  • Backend: Supabase (Postgres + Auth)
  • Storage: Supabase Storage
  • AI integration: Claude (Anthropic)
  • Deployment: Vercel

💥 Reality Bites

Week 1 was smooth. By Week 2, cracks appeared.

  • Supabase’s Postgres was fine… until our expensive queries choked it.
  • React Native was okay, but we hit prop-passing hell fast.
  • Claude integration was awesome… until we realized we needed more dynamic prompt chaining and Claude didn’t support that natively.
“Why is our dropdown breaking again?” — our frontender, twice a day.

🔄 The Course Correction

By Week 3, we knew we had to pivot — but only where it hurt most. We didn’t ditch everything. We changed enough to stay sane:

  • Moved login/email from Supabase to Claude + SendGrid for more flexibility
  • Switched Postgres → PlanetScale (MySQL-based, better handling of concurrent writes)
  • Replaced dropdowns and navigation logic with Radix + Expo Router

The app became 40% faster and way easier to debug.

📚 Lessons Burned into Our Brains

  • Speed > perfection: Don’t overplan the stack — plan for change.
  • Invest in dev experience early: Bad structure compounds pain fast.
  • Optimize for iteration, not architecture.

We still use Supabase in other projects. It just wasn’t right for this one at this stage.

⚖️ What Stayed

Some things did work well and stayed untouched:

  • Vercel + Edge Functions (fast deploys, perfect for MVP)
  • Claude for summaries + analysis (low latency, clean output)
  • Figma for quick UI updates

🎯 The Real Takeaway

“The right stack is the one that helps you ship fast — and makes you want to come back and improve it.”

🔜 Coming Up Next

Next Friday: Debugging Hell: One Bug That Nearly Crashed Our Launch

Labels:

Comments

Post a Comment

Popular posts from this blog

PromptCraft Blog Series #5: Automating Tasks With Prompt-Driven Workflows - Build AI-powered taskbots using no-code platforms like Lovable and Replit

PromptCraft Series #5 – Automating Tasks With Prompt Workflows ✨ PromptCraft Series #5 "Automating Tasks With Prompt-Driven Workflows" 🗕️ New post every Monday · Brought to you by Marc Rexian 🤖 Why Task Automation Matters With no-code platforms like Lovable and Replit , you can now build bots that: Summarize documents Generate reports Write replies Organize information Trigger API calls No Python. No cron jobs. Just prompts + flow. 🔧 What Is a Prompt-Driven Workflow? A user action or input starts the process A prompt block handles the logic The AI response is used to update the UI, send data, or trigger another action Think of it as Zapier powered by LLMs . ✨ TaskBot Use Cases You Can Build Today TaskBot Type Prompt Pattern Example ✉️ Email Writer ...
Post a Comment
Read more

PromptCraft Blog Series #6: Prompt Debugging and Optimization – Learn how to fix and improve AI prompt outputs for more accurate, helpful results.

PromptCraft Series #6 – Prompt Debugging and Optimization "As of May 2025, summarize one real, recent science discovery based on known sources. Add links if available and avoid speculation." ✨ PromptCraft Series #6 "Prompt Debugging and Optimization: Getting the Output You Want" 🗕️ New post every Monday 🔍 Why Prompts Sometimes Fail Even the best models can give you: ❌ Irrelevant answers ❌ Generic or vague responses ❌ Hallucinated facts or made-up data ❌ Wrong tone or misunderstanding of intent Often, it’s not the AI’s fault — it’s the prompt . 🔧 How to Debug a Prompt Start with these questions: Is the role or task clearly defined? Did you give examples or context? Are your constraints too loose or too strict? Did you format the output instructions properly? Then iterate your prompt, one element at...
Post a Comment
Read more

Behind the Scenes: How Generative AI Creates Music, Art, and Stories

When Machines Dream We’re living in a world where machines don’t just compute—they create. Generative AI is writing novels, composing symphonies, and painting pictures. But what’s really going on behind the screen? This post pulls back the curtain to reveal how generative AI actually creates —from writing a bedtime story to composing a lo-fi beat. Whether you're a curious creator or tech enthusiast, you’ll see the art of AI through a new lens. What is Generative AI, Really? Generative AI uses machine learning models—especially neural networks—to generate new content based on learned patterns. Trained on vast datasets, these models produce original music, images, and text based on user prompts. 1. How AI Writes Stories (e.g., ChatGPT, Claude) Step-by-step: You give a prompt: “Write a story about a lonely robot who finds a friend in the forest.” The model (like ChatGPT) draws on its training data to predict and generate the most likely next word, sentence, or paragr...
Post a Comment
Read more