How to build a 1.7M row consumer safety search engine on a 6GB RAM instance?

Hey Product Hunt community! 👋

I’m the solo builder behind LemonKnows, an independent consumer safety and food intelligence platform tracking over 1.7 million records, 41k global government recalls, and 100k+ complaints.

Because I wanted to keep the platform completely unbiased and self-funded, I challenged myself to build the entire infrastructure on a single 6GB RAM instance.

Under heavy data load, I hit massive latency bottlenecks. To solve them and achieve a blazing 66ms load speed, I had to get creative:

• Database Tuning: Rewrote heavy aggregations into custom stats cache tables, dropping COUNT queries from 3.3s down to 2ms.

• Zero-Cost AI Translations: Built a server-side translation proxy using Gemini and a SHA-256 text-hashed database lookup layer, letting us serve data in 7 languages natively without recurring API costs.

• Crowdsourced Image Loop: Designed a secure, auth-gated client-side submission pipeline so the community can easily suggest missing product photos into an approval queue.

Before our launch goes live on Tuesday, I’d love to open up the floor to other makers and developers.

I'd love your insights on:

1. What techniques do you use to keep your database footprints lean on free-tier or budget cloud servers?

2. How are you caching or optimizing LLM outputs to prevent API costs from spiraling?

Looking forward to trading ideas in the comments below! 🍋