Method queryHandler

ragserver/ragserver/main.go:124–192

(w http.ResponseWriter, req *http.Request)

}

func (rs *ragServer) queryHandler(w http.ResponseWriter, req *http.Request) {
	// Parse HTTP request from JSON.
	type queryRequest struct {
		Content string
	}
	qr := &queryRequest{}
	err := readRequestJSON(req, qr)
	if err != nil {
		http.Error(w, err.Error(), http.StatusBadRequest)
		return
	}

	// Embed the query contents.
	rsp, err := rs.embModel.EmbedContent(rs.ctx, genai.Text(qr.Content))
	if err != nil {
		http.Error(w, err.Error(), http.StatusInternalServerError)
		return
	}

	// Search weaviate to find the most relevant (closest in vector space)
	// documents to the query.
	gql := rs.wvClient.GraphQL()
	result, err := gql.Get().
		WithNearVector(
			gql.NearVectorArgBuilder().WithVector(rsp.Embedding.Values)).
		WithClassName("Document").
		WithFields(graphql.Field{Name: "text"}).
		WithLimit(3).
		Do(rs.ctx)
	if werr := combinedWeaviateError(result, err); werr != nil {
		http.Error(w, werr.Error(), http.StatusInternalServerError)
		return
	}

	contents, err := decodeGetResults(result)
	if err != nil {
		http.Error(w, fmt.Errorf("reading weaviate response: %w", err).Error(), http.StatusInternalServerError)
		return
	}

	// Create a RAG query for the LLM with the most relevant documents as
	// context.
	ragQuery := fmt.Sprintf(ragTemplateStr, qr.Content, strings.Join(contents, "\n"))
	resp, err := rs.genModel.GenerateContent(rs.ctx, genai.Text(ragQuery))
	if err != nil {
		log.Printf("calling generative model: %v", err.Error())
		http.Error(w, "generative model error", http.StatusInternalServerError)
		return
	}

	if len(resp.Candidates) != 1 {
		log.Printf("got %v candidates, expected 1", len(resp.Candidates))
		http.Error(w, "generative model error", http.StatusInternalServerError)
		return
	}

	var respTexts []string
	for _, part := range resp.Candidates[0].Content.Parts {