breakpilot-compliance/ai-compliance-sdk/internal/ucca/legal_rag_http.go

package ucca

import (
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"net/http"
)

// generateEmbedding calls Ollama bge-m3 to get a 1024-dim vector for the query.
func (c *LegalRAGClient) generateEmbedding(ctx context.Context, text string) ([]float64, error) {
	if len(text) > 2000 {
		text = text[:2000]
	}

	reqBody := ollamaEmbeddingRequest{
		Model:  c.embeddingModel,
		Prompt: text,
	}

	jsonBody, err := json.Marshal(reqBody)
	if err != nil {
		return nil, fmt.Errorf("failed to marshal embedding request: %w", err)
	}

	req, err := http.NewRequestWithContext(ctx, "POST", c.ollamaURL+"/api/embeddings", bytes.NewReader(jsonBody))
	if err != nil {
		return nil, fmt.Errorf("failed to create embedding request: %w", err)
	}
	req.Header.Set("Content-Type", "application/json")

	resp, err := c.httpClient.Do(req)
	if err != nil {
		return nil, fmt.Errorf("embedding request failed: %w", err)
	}
	defer resp.Body.Close()

	if resp.StatusCode != http.StatusOK {
		body, _ := io.ReadAll(resp.Body)
		return nil, fmt.Errorf("ollama returned %d: %s", resp.StatusCode, string(body))
	}

	var embResp ollamaEmbeddingResponse
	if err := json.NewDecoder(resp.Body).Decode(&embResp); err != nil {
		return nil, fmt.Errorf("failed to decode embedding response: %w", err)
	}

	if len(embResp.Embedding) == 0 {
		return nil, fmt.Errorf("no embedding returned from ollama")
	}

	return embResp.Embedding, nil
}

// ensureTextIndex creates a full-text index on chunk_text if not already done.
func (c *LegalRAGClient) ensureTextIndex(ctx context.Context, collection string) error {
	if c.textIndexEnsured[collection] {
		return nil
	}

	indexReq := qdrantTextIndexRequest{
		FieldName: "chunk_text",
		FieldSchema: qdrantTextFieldSchema{
			Type:      "text",
			Tokenizer: "word",
			MinLen:    2,
			MaxLen:    40,
		},
	}

	jsonBody, err := json.Marshal(indexReq)
	if err != nil {
		return fmt.Errorf("failed to marshal text index request: %w", err)
	}

	url := fmt.Sprintf("%s/collections/%s/index", c.qdrantURL, collection)
	req, err := http.NewRequestWithContext(ctx, "PUT", url, bytes.NewReader(jsonBody))
	if err != nil {
		return fmt.Errorf("failed to create text index request: %w", err)
	}
	req.Header.Set("Content-Type", "application/json")
	if c.qdrantAPIKey != "" {
		req.Header.Set("api-key", c.qdrantAPIKey)
	}

	resp, err := c.httpClient.Do(req)
	if err != nil {
		return fmt.Errorf("text index request failed: %w", err)
	}
	defer resp.Body.Close()

	// 200 = created, 409 = already exists — both are fine
	if resp.StatusCode != http.StatusOK && resp.StatusCode != http.StatusConflict {
		body, _ := io.ReadAll(resp.Body)
		return fmt.Errorf("text index creation failed %d: %s", resp.StatusCode, string(body))
	}

	c.textIndexEnsured[collection] = true
	return nil
}

// searchHybrid performs RRF-fused hybrid search (dense + full-text) via Qdrant Query API.
func (c *LegalRAGClient) searchHybrid(ctx context.Context, collection string, embedding []float64, regulationIDs []string, topK int) ([]qdrantSearchHit, error) {
	if err := c.ensureTextIndex(ctx, collection); err != nil {
		return nil, err
	}

	prefetchLimit := 20
	if topK > 20 {
		prefetchLimit = topK * 4
	}

	queryReq := qdrantQueryRequest{
		Prefetch: []qdrantPrefetch{
			{Query: embedding, Limit: prefetchLimit},
		},
		Query:       &qdrantFusion{Fusion: "rrf"},
		Limit:       topK,
		WithPayload: true,
	}

	if len(regulationIDs) > 0 {
		conditions := make([]qdrantCondition, len(regulationIDs))
		for i, regID := range regulationIDs {
			conditions[i] = qdrantCondition{
				Key:   "regulation_id",
				Match: qdrantMatch{Value: regID},
			}
		}
		queryReq.Filter = &qdrantFilter{Should: conditions}
	}

	jsonBody, err := json.Marshal(queryReq)
	if err != nil {
		return nil, fmt.Errorf("failed to marshal query request: %w", err)
	}

	url := fmt.Sprintf("%s/collections/%s/points/query", c.qdrantURL, collection)
	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(jsonBody))
	if err != nil {
		return nil, fmt.Errorf("failed to create query request: %w", err)
	}
	req.Header.Set("Content-Type", "application/json")
	if c.qdrantAPIKey != "" {
		req.Header.Set("api-key", c.qdrantAPIKey)
	}

	resp, err := c.httpClient.Do(req)
	if err != nil {
		return nil, fmt.Errorf("query request failed: %w", err)
	}
	defer resp.Body.Close()

	if resp.StatusCode != http.StatusOK {
		body, _ := io.ReadAll(resp.Body)
		return nil, fmt.Errorf("qdrant query returned %d: %s", resp.StatusCode, string(body))
	}

	var queryResp qdrantQueryResponse
	if err := json.NewDecoder(resp.Body).Decode(&queryResp); err != nil {
		return nil, fmt.Errorf("failed to decode query response: %w", err)
	}

	return queryResp.Result, nil
}

// searchDense performs a dense-only vector search via Qdrant /points/search.
func (c *LegalRAGClient) searchDense(ctx context.Context, collection string, embedding []float64, regulationIDs []string, topK int) ([]qdrantSearchHit, error) {
	searchReq := qdrantSearchRequest{
		Vector:      embedding,
		Limit:       topK,
		WithPayload: true,
	}

	if len(regulationIDs) > 0 {
		conditions := make([]qdrantCondition, len(regulationIDs))
		for i, regID := range regulationIDs {
			conditions[i] = qdrantCondition{
				Key:   "regulation_id",
				Match: qdrantMatch{Value: regID},
			}
		}
		searchReq.Filter = &qdrantFilter{Should: conditions}
	}

	return c.doPointsSearch(ctx, collection, searchReq)
}

// searchBinding fetches the top binding_law hits (authority-stratified pool) so the
// obligation source is always a candidate even when guidance dominates semantically.
// It AUGMENTS the semantic pool — guidance is preserved as interpretation context.
func (c *LegalRAGClient) searchBinding(ctx context.Context, collection string, embedding []float64, topK int) ([]qdrantSearchHit, error) {
	searchReq := qdrantSearchRequest{
		Vector:      embedding,
		Limit:       topK,
		WithPayload: true,
		Filter: &qdrantFilter{Must: []qdrantCondition{
			{Key: "source_class", Match: qdrantMatch{Value: "binding_law"}},
		}},
	}

	return c.doPointsSearch(ctx, collection, searchReq)
}

// doPointsSearch issues a POST /points/search and decodes the hits.
func (c *LegalRAGClient) doPointsSearch(ctx context.Context, collection string, searchReq qdrantSearchRequest) ([]qdrantSearchHit, error) {
	jsonBody, err := json.Marshal(searchReq)
	if err != nil {
		return nil, fmt.Errorf("failed to marshal search request: %w", err)
	}

	url := fmt.Sprintf("%s/collections/%s/points/search", c.qdrantURL, collection)
	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(jsonBody))
	if err != nil {
		return nil, fmt.Errorf("failed to create search request: %w", err)
	}
	req.Header.Set("Content-Type", "application/json")
	if c.qdrantAPIKey != "" {
		req.Header.Set("api-key", c.qdrantAPIKey)
	}

	resp, err := c.httpClient.Do(req)
	if err != nil {
		return nil, fmt.Errorf("search request failed: %w", err)
	}
	defer resp.Body.Close()

	if resp.StatusCode != http.StatusOK {
		body, _ := io.ReadAll(resp.Body)
		return nil, fmt.Errorf("qdrant returned %d: %s", resp.StatusCode, string(body))
	}

	var searchResp qdrantSearchResponse
	if err := json.NewDecoder(resp.Body).Decode(&searchResp); err != nil {
		return nil, fmt.Errorf("failed to decode search response: %w", err)
	}

	return searchResp.Result, nil
}