Find my slow queries and fix the worst ones. Enable slow query logging (or the platform's query performance view) with a threshold around 100ms, and seed realistic data volumes if dev data is tiny; queries that look fine on 20 rows are the ones that matter. For each query over the threshold: get the query plan (EXPLAIN or the platform equivalent); diagnose whether it needs an index, a rewrite, a smaller result set, or a different access pattern; apply the fix; and re-run the plan to confirm the improvement. Fix by evidence, not by sprinkling indexes everywhere: each new index slows down writes, so only add ones a plan justifies. If a query is inherently heavy (reports, exports), move it out of the request cycle instead of tuning it to death. Don't add indexes without a query plan showing they'll be used. Don't change what any query returns. Deliver: a table of the slow queries ranked by impact (frequency times duration), the plan before and after each fix, the fixes applied, and verification: each fixed query now under the threshold at realistic data volume.