Broblems is the product I wish existed during my last two failed startups. I built it after wasting 18 months on solutions nobody asked for.

The Business Angle

The concept is brutally simple: people complain constantly online, and those complaints are essentially market research being given away for free. We scrape this digital frustration, process it through LLMs, and transform it into structured problem data that founders and product teams can actually use.

What makes it viable:

• Low CAC through maker communities (Product Hunt, Indie Hackers)
• Credit-based business model with lifetime access (high perceived value)
• Extremely low operating costs once the initial scraping/processing pipeline is built
• Natural viral loop as builders share interesting problems they’ve discovered

The Technical Stack

The system architecture breaks down into three core components:

Data Acquisition Pipeline

• Distributed scrapers running on serverless functions (AWS Lambda)
• Reddit API integration with rotating tokens to avoid rate limiting
• Twitter/X API access for real-time frustration mining
• Custom puppeteer scripts for forums that block traditional scraping

Processing Engine

• NestJS backend with a modular architecture for handling different data sources
• PostgreSQL with JSON fields for flexible problem schema
• fine-tuned LLM for categorizing complaints as actionable problems
• Vector embeddings to group similar complaints and determine problem frequency
• Background workers for processing the scraping queue with Bull/Redis

Problem Discovery Interface

• React frontend with server components for SEO optimization
• SSE for real-time updates when new problems are discovered
• Credit-based gating for detailed reports

The trickiest technical challenge was distinguishing between actual product opportunities and general venting. We ended up creating a custom training dataset of 15,000 manually labeled social media complaints to fine-tune a gpt-3.5-turbo model.

at the start of the project, I was processing ~1000 social posts daily, identifying ~40-70 unique actionable problems each week across 32 different categories. Not bad for something I built because I was tired of building things nobody wanted.

The Money

The project is sitting at a 79$MRR, which is obviously not enough to quit my day job or pay my coffee bills. But it’s a fun side project that I’m still from time to time adding features to.