Solvely
Homework assistance

As the content library scaled, the legacy search system suffered from latency exceeding one second during peak usage hours. The previous provider's pricing model caused infrastructure costs to rise significantly faster than actual platform usage.

The team migrated to Zilliz Cloud to power a RAG pipeline that applies subject-specific normalization to preserve chemistry formulas and geometry diagrams. Student queries are processed through OCR and Google or OpenAI embedding models before executing vector similarity searches against the index. An LLM reranks the retrieved candidates to capture subtle context that vector search alone might miss.