I'm a senior data engineer with several years of experience, currently employed full-time, and I also have a US LLC. I’ve been thinking about building something on the side. No pressure to monetize quickly, since I don’t need immediate income. The idea I keep coming back to is offering AI services to SaaS/fintech companies that already tried to ship AI features and got stuck somewhere between “cool demo” and “usable product.” A pattern I keep seeing: * hallucinations make the system unreliable * nobody knows how to properly evaluate outputs * there’s no monitoring/feedback loop in production * prompts get tweaked endlessly without real measurement * the AI feature works 80% of the time, which means nobody fully trusts it My potential edge wouldn’t really be “AI agents.” It would be the data/reliability side: * evaluation pipelines * observability * grounding/retrieval * production monitoring * human feedback loops * data quality * making LLM systems actually dependable enough for business use What makes…