Why should you care?

Models hallucinate policy without grounding; RAG ties answers to tenant knowledge bases.

See it live — copy this example

Paste into an ASP.NET Core 8+ / AgentNest project, then run with dotnet run (set your API keys in user-secrets).

// AgentNest.Rag/RagPipeline.cs
public sealed class RagPipeline(IEmbeddingGenerator embedder, IVectorStore store, IChatClient chat)
{
    public async Task<string> AskAsync(string tenantId, string question, CancellationToken ct)
    {
        var vector = await embedder.GenerateEmbeddingAsync(question, cancellationToken: ct);
        var hits = await store.SearchAsync(tenantId, vector, topK: 5, ct);
        var context = string.Join("\n---\n", hits.Select(h => h.Text));
        var prompt = $"Use only this context:\n{context}\n\nQuestion: {question}";
        return (await chat.GetResponseAsync([prompt], cancellationToken: ct)).Text;
    }
}

What happened?

• RagPipeline embeds the question, searches tenant vectors, injects hits into the prompt, then generates the answer.
• Follow the steps below — typing the code yourself is the fastest way to learn.

Practice next

1. build IVectorStore abstraction in AgentNest.Rag.
2. Index CRM PDFs per tenant with embedding batch jobs.
3. Add RagPipeline to DI and expose POST /api/rag/ask.
4. Add reranker step after vector search.
5. Return citations array alongside answer text.

Remember

RAG = retrieve relevant chunks, then generate with context. Always filter vector search by tenantId. Log chunk IDs for traceability in AgentNest.

CRM playbook answers

Reps ask how to handle enterprise discount requests.

Outcome: RAG pulls approved playbook sections instead of inventing pricing policy.