Building an AI Resume-to-Startup Idea Generator with Claude Code

Last weekend, I decided to tackle a fun project: build a tool that transforms a plain-text resume into personalized startup ideas, complete with market insights, tech recommendations, and growth strategies. My goal was straightforward: showcase how effectively I can integrate AI into practical applications—and do it all within a weekend.

Here’s the journey of how it came together:

Why This Idea?

I often noticed friends and colleagues struggling to come up with actionable business ideas that leveraged their unique experiences. I thought:

“What if a resume could inspire real, tailored startup concepts?”

This would not only demonstrate my technical capabilities but also genuinely help people kickstart new ventures.

The Vision

Users submit their resumes.
They receive three well-thought-out startup concepts.
Ideas include market opportunities, target customers, recommended tech stacks, and actionable business strategies.
Keep the cost of running the AI low—under $0.05 per use.

Choosing the Tools

To keep things quick and effective, I picked familiar, robust technologies:

Astro for a fast, clean, SEO-friendly frontend.
AWS Lambda for a scalable and cost-efficient backend.
Claude Code for agentic, autonomous coding, enabling rapid prototyping and AI-assisted architecture decisions.

Finding the Right Model

Initially, I tried Amazon Bedrock with the DeepSeek-R1 model. While it generally performed well, I encountered occasional issues with malformed JSON outputs, requiring additional error handling and complicated parsing logic.

To resolve this, I switched to Amazon Bedrock with Sonnet-4. This significantly improved results, consistently producing clean and reliable JSON outputs. However, Sonnet-4’s superior performance came with higher costs, making it challenging to maintain the desired low cost per generation.

Ultimately, I discovered Google’s Gemini Flash 2.0, which provides an exceptionally generous free tier (2M tokens/day, effectively $0.00 per generation). Gemini Flash 2.0 offered the perfect balance of reliability and cost-efficiency, allowing me to deliver quality outputs consistently without overspending.

How This Was Built

Built with Agentic Coding Using Claude Code

Autonomous development using Claude Code for rapid prototyping
AI-assisted architecture decisions and implementation
Iterative refinement through human-AI collaboration

Technical Architecture

Frontend: Astro + TypeScript for optimal performance and SEO
Backend: AWS Lambda for serverless scalability
AI: Google Gemini Flash 2.0 free tier for cost-effective reasoning
Security: API token authentication for controlled access
Hosting: S3 static hosting with CloudFront CDN

Cost Optimization

Google Gemini Flash 2.0: Free tier with 2M tokens/day (~$0.00/generation)
Serverless architecture eliminates idle costs
Strategic prompt engineering minimizes API calls

Performance Features

Sub-30 second response times
Responsive design for all devices
Error handling and graceful degradation

Overcoming Challenges

One tricky part was handling inconsistent JSON outputs from the AI. After some creative problem-solving, I implemented robust parsing that gracefully manages unexpected formatting—ensuring users always get clear, actionable results.

Going Live

Deploying was straightforward: a few simple commands, and I was up and running, closely monitoring everything through AWS CloudWatch. API responses came back—typically just 10-15 seconds.