RAG guide
Better retrieval
If the right passage is not among the few sent to the model, the answer will be wrong or empty, however good the model is. Retrieval is where most of the tuning effort pays off.
SearchHybrid: meaning plus keywords
ThenRerank
Send to modelA few best passages
MeasureDid the right passage come back?
The choices that matter
| Choice | What it does | Our usual default |
|---|---|---|
| Hybrid search | Combines meaning-based (vector) search with keyword search | Always on. Keywords catch policy numbers, product codes and names that vector search misses. |
| Reranking | A second model re-scores the top results for relevance | On for most use cases. Often the single biggest quality gain. |
| Number of passages | How many passages the model sees | Start with 3 to 8, tuned on the test set |
| Metadata filters | Restrict by department, date, document type or permission | Use whenever the question implies a scope |
| Query rewriting | Turns a vague or follow-up question into a clear search | On for chat assistants with follow-up questions |
Languages
For users who ask in Hindi or mix Hindi and English, use a multilingual embedding model and test with real questions. Keyword search on transliterated text needs extra care.
More RAG guides: How RAG is built · Preparing documents · Permissions · Measuring answers · RAG FAQ · Use case: HR policy assistant
Want answers from your own documents?
Tell us where the documents live, who should be able to ask, and ten questions people ask today. We will come back with a plain view of what it takes to answer them well, and safely.