descrybe/apps/api/internal/processing/pinecone.go

package processing

import (
	"bytes"
	"context"
	"encoding/json"
	"errors"
	"io"
	"net/http"
	"strings"
	"time"
)

// PineconeCategorizer implements VectorCategorizer via Pinecone query API.
// When Embedder is set (admin AI role "vectorization" / env fallback), queries
// send an explicit vector; otherwise Text is used (Pinecone integrated inference).
// ASSUMPTION: when not configured, Enabled() is false and callers skip vector categorize.
type PineconeCategorizer struct {
	APIKey     string
	Host       string
	Namespace  string
	HTTPClient *http.Client
	Embedder   Embedder
}

func NewPineconeCategorizer(apiKey, host, namespace string) *PineconeCategorizer {
	return &PineconeCategorizer{
		APIKey:     strings.TrimSpace(apiKey),
		Host:       strings.TrimRight(strings.TrimSpace(host), "/"),
		Namespace:  namespace,
		HTTPClient: &http.Client{Timeout: 20 * time.Second},
	}
}

func (p *PineconeCategorizer) Enabled() bool {
	return p != nil && p.APIKey != "" && p.Host != ""
}

type pineconeQueryRequest struct {
	Namespace       string    `json:"namespace,omitempty"`
	TopK            int       `json:"topK"`
	IncludeMetadata bool      `json:"includeMetadata"`
	Vector          []float32 `json:"vector,omitempty"`
	Text            string    `json:"text,omitempty"`
}

type pineconeQueryResponse struct {
	Matches []struct {
		ID       string         `json:"id"`
		Score    float64        `json:"score"`
		Metadata map[string]any `json:"metadata"`
	} `json:"matches"`
}

func (p *PineconeCategorizer) SuggestCategory(ctx context.Context, companyID, productText string, candidates []string) (string, error) {
	if !p.Enabled() {
		return "", errors.New("pinecone not configured")
	}
	productText = SanitizeText(productText)
	if productText == "" {
		return "", errors.New("empty product text")
	}
	_ = companyID
	_ = candidates

	reqBody := pineconeQueryRequest{
		Namespace:       p.Namespace,
		TopK:            1,
		IncludeMetadata: true,
	}
	if p.Embedder != nil {
		vecs, err := p.Embedder.Embed(ctx, []string{productText})
		if err != nil {
			return "", err
		}
		if len(vecs) == 0 || len(vecs[0]) == 0 {
			return "", errors.New("empty embedding")
		}
		reqBody.Vector = vecs[0]
	} else {
		reqBody.Text = productText
	}

	body, err := json.Marshal(reqBody)
	if err != nil {
		return "", err
	}
	req, err := http.NewRequestWithContext(ctx, http.MethodPost, p.Host+"/query", bytes.NewReader(body))
	if err != nil {
		return "", err
	}
	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Api-Key", p.APIKey)

	res, err := p.HTTPClient.Do(req)
	if err != nil {
		return "", err
	}
	defer res.Body.Close()
	raw, err := io.ReadAll(io.LimitReader(res.Body, 1<<20))
	if err != nil {
		return "", err
	}
	if res.StatusCode >= 400 {
		return "", errors.New(TruncateError(errors.New("pinecone query failed")))
	}
	var parsed pineconeQueryResponse
	if err := json.Unmarshal(raw, &parsed); err != nil {
		return "", err
	}
	if len(parsed.Matches) == 0 {
		return "", errors.New("no pinecone matches")
	}
	m := parsed.Matches[0].Metadata
	if m != nil {
		if name, ok := m["category"].(string); ok && strings.TrimSpace(name) != "" {
			return SanitizeOutput(name), nil
		}
		if name, ok := m["name"].(string); ok && strings.TrimSpace(name) != "" {
			return SanitizeOutput(name), nil
		}
	}
	return SanitizeOutput(parsed.Matches[0].ID), nil
}

// NoopVectorCategorizer is the default when Pinecone is unset.
type NoopVectorCategorizer struct{}

func (NoopVectorCategorizer) Enabled() bool { return false }

func (NoopVectorCategorizer) SuggestCategory(context.Context, string, string, []string) (string, error) {
	return "", errors.New("vector categorizer disabled")
}