Building AI apps

Hi there! 👋

This week I shipped Specsage using CrewAI and it's been receiving great feedback from the community. I wanted to share what I learned building an AI-powered application.

What is Specsage?

Specsage is an AI-powered PC builder assistant. It helps users find the perfect PC configuration based on their needs and budget using AI agents.

Live: specsage.tech

Key Learnings

1. CrewAI agents are incredibly powerful

CrewAI let me create specialized agents that work together. I had:

A research agent that gathers PC specs
An analysis agent that evaluates value
A recommendation agent that provides final suggestions

The agents work together seamlessly, and the results are surprisingly good.

2. Prompting matters more than code

I spent 40% of the time getting prompts right. Better prompts = better agent behavior. A well-crafted system prompt is worth 10x the code optimization.

3. Firecrawl saves so much time

Instead of building web scrapers, I used Firecrawl to extract PC specifications from multiple sites. It's a game-changer for data collection.

What's Next?

I'm currently exploring:

Implementing multi-turn conversations
Adding user preference learning
Building a feedback loop to improve recommendations

Want to hear about my next projects? Subscribe below to get updates when I ship something new.

Owen
Building with Next.js, AI, and anything that solves problems.