From 3fa3561bb2311df143c84984e0a4bcfd4a80e498 Mon Sep 17 00:00:00 2001 From: dindang Date: Mon, 27 Jul 2026 15:34:32 +0800 Subject: [PATCH] =?UTF-8?q?=E9=97=AE=E5=8D=B7?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .env.example | 25 + clip_server/server.py | 90 +++ internal/api/index.html | 440 +++++++++++++ internal/api/index_handler.go | 10 + internal/api/index_html.go | 3 + internal/api/questionnaire.html | 359 +++++++++++ internal/api/router.go | 5 + internal/handler/questionnaire_handler.go | 713 ++++++++++++++++++++++ internal/repository/product_repo.go | 9 + 9 files changed, 1654 insertions(+) create mode 100644 .env.example create mode 100644 clip_server/server.py create mode 100644 internal/api/index.html create mode 100644 internal/api/questionnaire.html create mode 100644 internal/handler/questionnaire_handler.go diff --git a/.env.example b/.env.example new file mode 100644 index 0000000..66470e8 --- /dev/null +++ b/.env.example @@ -0,0 +1,25 @@ +# ─── 服务器 ─── +PORT=8099 + +# ─── 数据库 ─── +DATABASE_URL=postgres://postgres:your_password@localhost:5432/floorvisualizer?sslmode=disable + +# ─── Redis ─── +REDIS_ADDR=localhost:6379 + +# ─── JWT ─── +JWT_SECRET=change-this-to-a-strong-secret + +# ─── AI ─── +OPENROUTER_API_KEY=sk-or-v1-your-openrouter-key-here + +# ─── 代理（国内环境用，留空则直连） ─── +PROXY_URL=http://127.0.0.1:7897 +# DISABLE_PROXY=true + +# ─── 队列 ─── +WORKER_COUNT=5 + +# ─── 日志 ─── +LOG_DIR=logs +LOG_LEVEL=INFO diff --git a/clip_server/server.py b/clip_server/server.py new file mode 100644 index 0000000..57c3bdb --- /dev/null +++ b/clip_server/server.py @@ -0,0 +1,90 @@ +"""CLIP-based indoor room classifier. Runs ViT-B/32 on CPU, ~100ms per image.""" + +import io +import logging +import os + +import open_clip +import torch +from PIL import Image +from flask import Flask, jsonify, request + +logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s") +log = logging.getLogger("clip-server") +app = Flask(__name__) +app.config["MAX_CONTENT_LENGTH"] = 50 * 1024 * 1024 # 50 MB max upload + +# Load once at startup +MODEL_NAME = os.getenv("CLIP_MODEL", "ViT-B-32") +PRETRAINED = os.getenv("CLIP_PRETRAINED", "laion2b_s34b_b79k") +DEVICE = "cuda" if torch.cuda.is_available() else "cpu" + +log.info(f"Loading CLIP model: {MODEL_NAME} pretrained={PRETRAINED} device={DEVICE}") +model, _, preprocess = open_clip.create_model_and_transforms(MODEL_NAME, pretrained=PRETRAINED) +model = model.to(DEVICE).eval() +tokenizer = open_clip.get_tokenizer(MODEL_NAME) + +# Text prompts for indoor/outdoor classification +PROMPTS = [ + "a photo of an indoor room interior with floor visible", + "a photo of an outdoor scene, landscape, street, or non-room image", +] + + +@app.route("/health", methods=["GET"]) +def health(): + return jsonify({"status": "ok", "model": MODEL_NAME, "device": DEVICE}) + + +@app.route("/check", methods=["POST"]) +def check(): + """Classify whether an image is an indoor room photo suitable for floor replacement. + Expects raw image bytes as the request body. Returns {score: 0-10, passed: bool}.""" + if not request.data: + return jsonify({"error": "no image data"}), 400 + + try: + img = Image.open(io.BytesIO(request.data)).convert("RGB") + except Exception as e: + return jsonify({"error": f"bad image: {e}"}), 400 + + # Multiple prompt pairs for robust ensemble (indoor vs outdoor) + prompt_pairs = [ + ("a photo of an indoor room interior with floor", "a photo of an outdoor scene"), + ("interior room photograph, real estate photo", "exterior photograph, outdoor photo"), + ("an indoor space with visible flooring", "an outdoor space, landscape, street view"), + ] + + image_input = preprocess(img).unsqueeze(0).to(DEVICE) + wins = 0 + gap_sum = 0.0 + for indoor_text, outdoor_text in prompt_pairs: + text_input = tokenizer([indoor_text, outdoor_text]).to(DEVICE) + with torch.no_grad(): + image_features = model.encode_image(image_input) + text_features = model.encode_text(text_input) + image_features /= image_features.norm(dim=-1, keepdim=True) + text_features /= text_features.norm(dim=-1, keepdim=True) + raw = (image_features @ text_features.T)[0] # cosine similarities + indoor_sim = float(raw[0].item()) + outdoor_sim = float(raw[1].item()) + if indoor_sim > outdoor_sim: + wins += 1 + gap_sum += indoor_sim - outdoor_sim # positive = indoor wins + + indoor_wins = wins >= 2 # majority vote (2 out of 3) + avg_gap = gap_sum / len(prompt_pairs) + # Map avg_gap to 1-10: gap ~0.05 = score 6, gap ~0.15+ = score 10 + score_1_10 = max(1, min(10, int(5 + avg_gap * 40))) + passed = indoor_wins + + log.info( + f"wins={wins}/3 avg_gap={avg_gap:.3f} score_1_10={score_1_10} passed={passed}" + ) + return jsonify({"score": score_1_10, "passed": passed}) + + +if __name__ == "__main__": + port = int(os.getenv("CLIP_PORT", "5100")) + log.info(f"CLIP server listening on :{port}") + app.run(host="0.0.0.0", port=port, debug=False) diff --git a/internal/api/index.html b/internal/api/index.html new file mode 100644 index 0000000..bc93fe8 --- /dev/null +++ b/internal/api/index.html @@ -0,0 +1,440 @@ + + + + + +FloorVisualizer Demo + + + +
+

FloorVisualizer AI 换地板 Demo

+

上传房间照片，选择真实产品、铺设方式和可选效果，生成写实换地板预览

+
+ +
+
+
+

第一步：上传房间照片

+
+ + + 点击或拖拽上传图片
建议使用能清楚看到地面的室内照片 +
+
+ +
+

房间类型

+
+
+ +
+

地板类型

+
+
木地板
Hardwood
+
瓷砖
Tile / Stone
+
乙烯基地板
Luxury Vinyl / SPC
+
特殊测试
7款 · 含AI描述
+
+
+ +
+

铺设方式

+
+
+ + + +
+

选择地板款式

+
+
+
+ +
+

材质参考方式

+
+
仅图片
参考图
+
仅JSON
材质描述
+
图片+JSON
两者都用
+
+
+ +
+

可选效果

+
+ + +
+
+ + +
+
+
+ +
+
+
+

生成失败原因

+ 

+    

+  

+
+  

+    
生成结果

+    
上传图片并选择地板选项后
点击生成预览图查看效果

+    生成结果
+    

+      生成图
+      
原图

+      
↔

+      
原图生成图

+    

+  

+

+
+
+
+
diff --git a/internal/api/index_handler.go b/internal/api/index_handler.go
index 67df1db..02d6e64 100644
--- a/internal/api/index_handler.go
+++ b/internal/api/index_handler.go
@@ -11,3 +11,13 @@ func Index(w http.ResponseWriter, r *http.Request) {
 	w.WriteHeader(http.StatusOK)
 	_, _ = w.Write([]byte(indexHTML))
 }
+
+func Questionnaire(w http.ResponseWriter, r *http.Request) {
+	if r.URL.Path != "/questionnaire" {
+		http.NotFound(w, r)
+		return
+	}
+	w.Header().Set("Content-Type", "text/html; charset=utf-8")
+	w.WriteHeader(http.StatusOK)
+	_, _ = w.Write([]byte(questionnaireHTML))
+}
diff --git a/internal/api/index_html.go b/internal/api/index_html.go
index bc888e3..f700edb 100644
--- a/internal/api/index_html.go
+++ b/internal/api/index_html.go
@@ -4,3 +4,6 @@ import _ "embed"
 
 //go:embed index.html
 var indexHTML string
+
+//go:embed questionnaire.html
+var questionnaireHTML string
diff --git a/internal/api/questionnaire.html b/internal/api/questionnaire.html
new file mode 100644
index 0000000..fa8a64d
--- /dev/null
+++ b/internal/api/questionnaire.html
@@ -0,0 +1,359 @@
+
+
+
+
+
+地板偏好问卷 - FloorVisualizer
+
+
+
+

+  
+
+  

+    

+      

+        
问卷推荐结果

+        
回答左侧核心问题后，系统会对产品库打分、合并同款不同尺寸，并返回最匹配的 10 款。

+      

+      

+        
-候选 SKU

+        
-去重款式

+        
0%问卷完成度

+      

+    

+    

+    
请选择你的空间、风格、外观、颜色和预算。

+    

+  

+

+
+
+
+
diff --git a/internal/api/router.go b/internal/api/router.go
index 7d942f8..238ac88 100644
--- a/internal/api/router.go
+++ b/internal/api/router.go
@@ -32,6 +32,8 @@ func RegisterRoutes(mux *http.ServeMux, client *openrouter.Client, db *sql.DB, a
 		productOpts      = handler.ProductOptions(db)
 		filterOpts       = handler.FilterOptions(db)
 		recommend        = handler.Recommend(db)
+		questionnaireRec = handler.QuestionnaireRecommend(db)
+		questionnaireOpt = handler.QuestionnaireOptions(db)
 		floorGenerate    = handler.FloorGenerate(client, db)
 		floorOptions     = handler.FloorOptions(db)
 		articleRecommend = handler.ArticleRecommend(db)
@@ -56,6 +58,9 @@ func RegisterRoutes(mux *http.ServeMux, client *openrouter.Client, db *sql.DB, a
 	mux.HandleFunc("/product-options", productOpts)
 	mux.HandleFunc("/filter-options", filterOpts)
 	mux.HandleFunc("/recommend/api", recommend)
+	mux.HandleFunc("/questionnaire", Questionnaire)
+	mux.HandleFunc("/questionnaire/options", questionnaireOpt)
+	mux.HandleFunc("/questionnaire/recommend", questionnaireRec)
 	mux.HandleFunc("/calculator/calc", handler.Calc)
 	mux.HandleFunc("/floor/generate", floorGenerate)
 	mux.HandleFunc("/floor/status", handler.FloorStatus)
diff --git a/internal/handler/questionnaire_handler.go b/internal/handler/questionnaire_handler.go
new file mode 100644
index 0000000..8ac90db
--- /dev/null
+++ b/internal/handler/questionnaire_handler.go
@@ -0,0 +1,713 @@
+package handler
+
+import (
+	"database/sql"
+	"encoding/json"
+	"fmt"
+	"math"
+	"net/http"
+	"sort"
+	"strings"
+
+	"floorvisualizer/internal/model"
+	"floorvisualizer/internal/repository"
+)
+
+type QuestionnaireBudget struct {
+	Min float64 `json:"min"`
+	Max float64 `json:"max"`
+}
+
+type QuestionnaireRequest struct {
+	Rooms          []string            `json:"rooms"`
+	Styles         []string            `json:"styles"`
+	Looks          []string            `json:"looks"`
+	ColorTones     []string            `json:"color_tones"`
+	Budget         QuestionnaireBudget `json:"budget"`
+	Brands         []string            `json:"brands"`
+	BrandMode      string              `json:"brand_mode"`
+	Performance    []string            `json:"performance"`
+	Finishes       []string            `json:"finishes"`
+	SizePreference string              `json:"size_preference"`
+	Priority       string              `json:"priority"`
+	Answers        map[string]any      `json:"answers,omitempty"`
+	Limit          int                 `json:"limit"`
+}
+
+type QuestionnaireRecommendation struct {
+	Product   model.Product `json:"product"`
+	Score     float64       `json:"score"`
+	Reasons   []string      `json:"reasons"`
+	SpecCount int           `json:"spec_count"`
+}
+
+type QuestionnaireResponse struct {
+	Recommendations  []QuestionnaireRecommendation `json:"recommendations"`
+	TotalCandidates  int                           `json:"total_candidates"`
+	UniqueCandidates int                           `json:"unique_candidates"`
+	AnsweredWeights  float64                       `json:"answered_weights"`
+}
+
+type questionnaireScoredProduct struct {
+	product model.Product
+	score   float64
+	reasons []string
+}
+
+func QuestionnaireOptions(db *sql.DB) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		if db == nil {
+			writeErr(w, http.StatusServiceUnavailable, "database is not available")
+			return
+		}
+		out := map[string]any{
+			"brands":      queryOptionCounts(db, "brand", 20),
+			"categories":  queryOptionCounts(db, "category", 20),
+			"materials":   queryOptionCounts(db, "material", 40),
+			"color_tones": queryOptionCounts(db, "color_tone", 30),
+			"finishes":    queryOptionCounts(db, "finish", 20),
+			"price_stats": queryPriceStats(db),
+		}
+		writeJSON(w, http.StatusOK, out)
+	}
+}
+
+func QuestionnaireRecommend(db *sql.DB) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		if db == nil {
+			writeErr(w, http.StatusServiceUnavailable, "database is not available")
+			return
+		}
+		if r.Method != http.MethodPost {
+			writeErr(w, http.StatusMethodNotAllowed, "method not allowed")
+			return
+		}
+
+		var req QuestionnaireRequest
+		if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+			writeErr(w, http.StatusBadRequest, "invalid questionnaire payload")
+			return
+		}
+		normalizeQuestionnaireRequest(&req)
+		if req.Limit <= 0 || req.Limit > 30 {
+			req.Limit = 10
+		}
+
+		products, err := repository.QueryAllProductRows(db)
+		if err != nil {
+			writeErr(w, http.StatusInternalServerError, err.Error())
+			return
+		}
+
+		brandOnly := req.BrandMode == "only" && len(req.Brands) > 0
+		brandSet := toSet(req.Brands)
+		scored := make([]questionnaireScoredProduct, 0, len(products))
+		for _, p := range products {
+			if !recommendableProduct(p) {
+				continue
+			}
+			if brandOnly && !brandSet[p.Brand] {
+				continue
+			}
+			score, reasons, answeredWeight := scoreQuestionnaireProduct(req, p)
+			if answeredWeight == 0 {
+				score = neutralProductScore(p)
+			}
+			scored = append(scored, questionnaireScoredProduct{product: p, score: roundOne(score), reasons: reasons})
+		}
+
+		unique := dedupeQuestionnaireResults(scored)
+		sort.SliceStable(unique, func(i, j int) bool {
+			if unique[i].score != unique[j].score {
+				return unique[i].score > unique[j].score
+			}
+			if unique[i].product.IsOfficialPrice != unique[j].product.IsOfficialPrice {
+				return unique[i].product.IsOfficialPrice
+			}
+			if req.Priority == "budget" && unique[i].product.PricePerSqft != unique[j].product.PricePerSqft {
+				return unique[i].product.PricePerSqft < unique[j].product.PricePerSqft
+			}
+			return unique[i].product.MainImageURL != "" && unique[j].product.MainImageURL == ""
+		})
+
+		selected := diversifyQuestionnaireResults(unique, req.Limit, brandOnly)
+		skus := make([]string, 0, len(selected))
+		for _, r := range selected {
+			skus = append(skus, r.product.SKU)
+		}
+		specCounts := repository.GetSpecCounts(db, skus)
+
+		recs := make([]QuestionnaireRecommendation, 0, len(selected))
+		for _, r := range selected {
+			key := r.product.Brand + "|" + r.product.StyleName + "|" + r.product.SeriesName
+			recs = append(recs, QuestionnaireRecommendation{
+				Product:   r.product,
+				Score:     r.score,
+				Reasons:   fallbackReasons(r.reasons, r.product),
+				SpecCount: specCounts[key],
+			})
+		}
+
+		_, _, answeredWeight := scoreQuestionnaireProduct(req, model.Product{})
+		writeJSON(w, http.StatusOK, QuestionnaireResponse{
+			Recommendations:  recs,
+			TotalCandidates:  len(scored),
+			UniqueCandidates: len(unique),
+			AnsweredWeights:  roundOne(answeredWeight),
+		})
+	}
+}
+
+func normalizeQuestionnaireRequest(req *QuestionnaireRequest) {
+	req.Rooms = cleanList(req.Rooms)
+	req.Styles = cleanList(req.Styles)
+	req.Looks = cleanList(req.Looks)
+	req.ColorTones = cleanList(req.ColorTones)
+	req.Brands = cleanList(req.Brands)
+	req.Performance = cleanList(req.Performance)
+	req.Finishes = cleanList(req.Finishes)
+	req.SizePreference = strings.TrimSpace(req.SizePreference)
+	req.Priority = strings.TrimSpace(req.Priority)
+	req.BrandMode = strings.TrimSpace(req.BrandMode)
+}
+
+func scoreQuestionnaireProduct(req QuestionnaireRequest, p model.Product) (float64, []string, float64) {
+	weights := map[string]float64{
+		"color":       20,
+		"style":       20,
+		"look":        15,
+		"budget":      15,
+		"room":        10,
+		"performance": 10,
+		"brand":       7,
+		"finish":      5,
+		"size":        3,
+	}
+	adjustWeights(weights, req.Priority)
+
+	totalWeight := 0.0
+	weighted := 0.0
+	reasons := []string{}
+
+	add := func(name string, answered bool, score float64, reason string) {
+		if !answered {
+			return
+		}
+		w := weights[name]
+		totalWeight += w
+		weighted += clampScore(score) * w
+		if score >= 78 && reason != "" {
+			reasons = append(reasons, reason)
+		}
+	}
+
+	add("color", len(req.ColorTones) > 0, scoreExactOrRelated(p.ColorTone, req.ColorTones, colorRelatedGroups()), fmt.Sprintf("颜色接近 %s", p.ColorTone))
+	add("style", len(req.Styles) > 0, scoreStyles(p, req.Styles), "风格偏好匹配")
+	add("look", len(req.Looks) > 0, scoreLooks(p, req.Looks), "外观类型匹配")
+	add("budget", req.Budget.Min > 0 || req.Budget.Max > 0, scoreBudget(p.PricePerSqft, req.Budget), priceReason(p))
+	add("room", len(req.Rooms) > 0, scoreRooms(p, req.Rooms), "适合选择的使用空间")
+	add("performance", len(req.Performance) > 0, scorePerformance(p, req.Performance), "性能偏好匹配")
+	add("brand", len(req.Brands) > 0 && req.BrandMode != "only", scoreExactOrRelated(p.Brand, req.Brands, nil), fmt.Sprintf("匹配品牌 %s", p.Brand))
+	add("finish", len(req.Finishes) > 0, scoreExactOrRelated(p.Finish, req.Finishes, [][]string{{"Matte", "Textured"}, {"Distressed/Embossed", "Wire Brushed"}}), fmt.Sprintf("表面质感为 %s", p.Finish))
+	add("size", req.SizePreference != "", scoreSize(p, req.SizePreference), "规格视觉匹配")
+
+	if totalWeight == 0 {
+		return 0, reasons, 0
+	}
+
+	score := weighted / totalWeight
+	if p.MainImageURL != "" || p.RoomImageURL != "" {
+		score += 1.5
+	}
+	if p.IsOfficialPrice {
+		score += 1
+	}
+	return clampScore(score), compactReasons(reasons), totalWeight
+}
+
+func adjustWeights(weights map[string]float64, priority string) {
+	switch priority {
+	case "style":
+		weights["style"] += 8
+		weights["color"] += 4
+	case "budget":
+		weights["budget"] += 12
+	case "durability":
+		weights["performance"] += 10
+		weights["room"] += 4
+	case "brand":
+		weights["brand"] += 12
+	case "color":
+		weights["color"] += 12
+	}
+}
+
+func recommendableProduct(p model.Product) bool {
+	status := strings.ToLower(strings.TrimSpace(p.Status))
+	if strings.Contains(status, "discontinued") || strings.Contains(status, "inactive") || strings.Contains(status, "下架") {
+		return false
+	}
+	return p.SKU != "" && p.StyleName != ""
+}
+
+func neutralProductScore(p model.Product) float64 {
+	score := 60.0
+	if p.MainImageURL != "" || p.RoomImageURL != "" {
+		score += 8
+	}
+	if p.PricePerSqft > 0 {
+		score += 5
+	}
+	if p.IsOfficialPrice {
+		score += 3
+	}
+	return score
+}
+
+func scoreExactOrRelated(value string, selected []string, related [][]string) float64 {
+	if value == "" || len(selected) == 0 {
+		return 50
+	}
+	for _, s := range selected {
+		if strings.EqualFold(value, s) {
+			return 100
+		}
+	}
+	for _, group := range related {
+		if containsFold(group, value) {
+			for _, s := range selected {
+				if containsFold(group, s) {
+					return 70
+				}
+			}
+		}
+	}
+	return 15
+}
+
+func colorRelatedGroups() [][]string {
+	return [][]string{
+		{"Natural", "Warm Yellow", "Taupe"},
+		{"Light Gray", "Off White", "Taupe"},
+		{"Dark Brown", "Black Walnut", "Cherry"},
+		{"Green", "Blue", "Light Gray"},
+	}
+}
+
+func scoreStyles(p model.Product, styles []string) float64 {
+	best := 0.0
+	for _, style := range styles {
+		best = math.Max(best, scoreStyle(p, style))
+	}
+	return best
+}
+
+func scoreStyle(p model.Product, style string) float64 {
+	text := productText(p)
+	switch style {
+	case "modern":
+		return tagScore(p, []string{"Light Gray", "Off White", "Natural"}, []string{"Maple", "Ash", "Porcelain"}, []string{"modern", "clean", "minimal", "airy", "sophisticated"}, text)
+	case "warm_natural":
+		return tagScore(p, []string{"Natural", "Warm Yellow", "Taupe"}, []string{"Oak", "Hickory", "Pine", "Ash"}, []string{"warm", "natural", "wood", "honey"}, text)
+	case "classic":
+		return tagScore(p, []string{"Cherry", "Dark Brown", "Black Walnut", "Natural"}, []string{"Oak", "Walnut", "Maple", "Hickory"}, []string{"classic", "traditional", "timeless"}, text)
+	case "rustic":
+		score := tagScore(p, []string{"Natural", "Dark Brown", "Taupe"}, []string{"Hickory", "Oak", "Pine"}, []string{"rustic", "scraped", "weathered", "reclaimed", "vintage", "barn"}, text)
+		if p.Finish == "Distressed/Embossed" || p.Finish == "Wire Brushed" {
+			score += 20
+		}
+		return clampScore(score)
+	case "luxury_stone":
+		score := tagScore(p, []string{"Off White", "Light Gray", "Natural"}, []string{"Porcelain", "Stone Look"}, []string{"marble", "travertine", "polished", "stone", "luxe"}, text)
+		if p.Category == "Tile/Stone" {
+			score += 25
+		}
+		if p.Finish == "Gloss" {
+			score += 10
+		}
+		return clampScore(score)
+	case "industrial":
+		return tagScore(p, []string{"Light Gray", "Black Walnut", "Dark Brown"}, []string{"Porcelain", "Stone Look"}, []string{"concrete", "slate", "smoke", "graphite", "cement", "industrial"}, text)
+	case "coastal":
+		return tagScore(p, []string{"Off White", "Light Gray", "Natural"}, []string{"Oak", "Maple", "Ash"}, []string{"beach", "coastal", "sand", "airy", "white"}, text)
+	}
+	return 50
+}
+
+func tagScore(p model.Product, colors, materials, keywords []string, text string) float64 {
+	score := 15.0
+	if containsFold(colors, p.ColorTone) {
+		score += 30
+	}
+	if containsFold(materials, p.Material) || materialFamilyHit(p.Material, materials) {
+		score += 25
+	}
+	for _, kw := range keywords {
+		if strings.Contains(text, kw) {
+			score += 15
+			break
+		}
+	}
+	if p.Finish == "Matte" {
+		score += 8
+	}
+	return clampScore(score)
+}
+
+func scoreLooks(p model.Product, looks []string) float64 {
+	best := 0.0
+	text := productText(p)
+	for _, look := range looks {
+		score := 0.0
+		switch look {
+		case "wood_look":
+			if isWoodLike(p, text) {
+				score = 100
+			} else if p.Category == "SPC-LVP" || p.Category == "Laminate" {
+				score = 70
+			} else {
+				score = 20
+			}
+		case "stone_tile":
+			if p.Category == "Tile/Stone" || strings.Contains(text, "stone") || strings.Contains(text, "marble") || strings.Contains(text, "travertine") {
+				score = 100
+			} else {
+				score = 15
+			}
+		case "real_hardwood":
+			if p.Category == "Solid Hardwood" || p.Category == "Engineered Hardwood" {
+				score = 100
+			} else if isWoodLike(p, text) {
+				score = 55
+			} else {
+				score = 10
+			}
+		case "light_clean":
+			score = scoreExactOrRelated(p.ColorTone, []string{"Off White", "Light Gray", "Natural"}, colorRelatedGroups())
+		case "dark_rich":
+			score = scoreExactOrRelated(p.ColorTone, []string{"Dark Brown", "Black Walnut", "Cherry"}, colorRelatedGroups())
+		}
+		best = math.Max(best, score)
+	}
+	return best
+}
+
+func scoreBudget(price float64, budget QuestionnaireBudget) float64 {
+	if price <= 0 {
+		return 50
+	}
+	min, max := budget.Min, budget.Max
+	if min > 0 && price < min {
+		return clampScore(100 - ((min-price)/math.Max(min, 1))*80)
+	}
+	if max > 0 && price > max {
+		return clampScore(100 - ((price-max)/math.Max(max, 1))*90)
+	}
+	return 100
+}
+
+func scoreRooms(p model.Product, rooms []string) float64 {
+	best := 0.0
+	for _, room := range rooms {
+		score := 50.0
+		switch room {
+		case "bathroom", "kitchen", "basement", "commercial":
+			score = scorePerformance(p, []string{"waterproof", "easy_clean"})
+		case "living_room", "bedroom":
+			score = math.Max(scoreStyles(p, []string{"warm_natural", "modern"}), 60)
+		case "rental":
+			score = scorePerformance(p, []string{"scratch_resistant", "easy_clean"})
+		}
+		best = math.Max(best, score)
+	}
+	return best
+}
+
+func scorePerformance(p model.Product, performance []string) float64 {
+	best := 0.0
+	text := productText(p)
+	for _, perf := range performance {
+		score := 20.0
+		switch perf {
+		case "waterproof":
+			if p.Category == "Tile/Stone" || p.Category == "SPC-LVP" || containsAnyFold(p.Material, []string{"porcelain", "vinyl", "rigid core", "spc"}) || strings.Contains(text, "water") || strings.Contains(text, "moisture") {
+				score = 100
+			}
+		case "scratch_resistant":
+			if p.Category == "Tile/Stone" || p.Category == "Laminate" || p.Category == "SPC-LVP" || containsAnyFold(text, []string{"scratch", "durable", "superguard", "wear"}) {
+				score = 95
+			}
+		case "easy_clean":
+			if p.Category == "Tile/Stone" || p.Category == "SPC-LVP" || p.Finish == "Matte" || containsAnyFold(text, []string{"clean", "maintenance"}) {
+				score = 90
+			}
+		case "pet_kid":
+			score = math.Max(scorePerformance(p, []string{"waterproof"}), scorePerformance(p, []string{"scratch_resistant"}))
+		case "comfort_quiet":
+			if p.Brand == "COREtec" || p.Category == "SPC-LVP" || containsAnyFold(text, []string{"quiet", "comfort", "soft", "underlayment"}) {
+				score = 90
+			}
+		case "eco":
+			if containsAnyFold(text, []string{"recycled", "eco", "green", "low voc"}) {
+				score = 90
+			} else {
+				score = 45
+			}
+		}
+		best = math.Max(best, score)
+	}
+	return best
+}
+
+func scoreSize(p model.Product, pref string) float64 {
+	switch pref {
+	case "wide_plank":
+		if p.WidthIn >= 7 && p.LengthIn >= 48 {
+			return 100
+		}
+		if p.WidthIn >= 5 {
+			return 70
+		}
+	case "standard_plank":
+		if p.WidthIn >= 4 && p.WidthIn < 7 && p.LengthIn >= 36 {
+			return 100
+		}
+		return 55
+	case "large_tile":
+		if p.Category == "Tile/Stone" && p.WidthIn >= 12 && p.LengthIn >= 24 {
+			return 100
+		}
+		if p.Category == "Tile/Stone" {
+			return 65
+		}
+	case "small_tile":
+		if p.Category == "Tile/Stone" && (p.WidthIn <= 6 || p.LengthIn <= 12) {
+			return 100
+		}
+		if p.Category == "Tile/Stone" {
+			return 60
+		}
+	}
+	return 20
+}
+
+func dedupeQuestionnaireResults(results []questionnaireScoredProduct) []questionnaireScoredProduct {
+	best := map[string]questionnaireScoredProduct{}
+	for _, r := range results {
+		key := strings.ToLower(r.product.Brand + "|" + r.product.SeriesName + "|" + r.product.StyleName + "|" + r.product.ColorTone)
+		if old, ok := best[key]; !ok || r.score > old.score || (r.score == old.score && r.product.MainImageURL != "" && old.product.MainImageURL == "") {
+			best[key] = r
+		}
+	}
+	out := make([]questionnaireScoredProduct, 0, len(best))
+	for _, r := range best {
+		out = append(out, r)
+	}
+	return out
+}
+
+func diversifyQuestionnaireResults(results []questionnaireScoredProduct, limit int, brandOnly bool) []questionnaireScoredProduct {
+	if len(results) <= limit {
+		return results
+	}
+	maxPerBrand := 4
+	if brandOnly {
+		maxPerBrand = limit
+	}
+	brandCounts := map[string]int{}
+	selected := make([]questionnaireScoredProduct, 0, limit)
+	for _, r := range results {
+		if brandCounts[r.product.Brand] >= maxPerBrand {
+			continue
+		}
+		selected = append(selected, r)
+		brandCounts[r.product.Brand]++
+		if len(selected) == limit {
+			return selected
+		}
+	}
+	for _, r := range results {
+		exists := false
+		for _, s := range selected {
+			if s.product.SKU == r.product.SKU {
+				exists = true
+				break
+			}
+		}
+		if !exists {
+			selected = append(selected, r)
+		}
+		if len(selected) == limit {
+			break
+		}
+	}
+	return selected
+}
+
+func fallbackReasons(reasons []string, p model.Product) []string {
+	reasons = compactReasons(reasons)
+	if len(reasons) > 0 {
+		return reasons
+	}
+	out := []string{}
+	if p.ColorTone != "" {
+		out = append(out, "颜色为 "+p.ColorTone)
+	}
+	if p.Material != "" {
+		out = append(out, "材质为 "+p.Material)
+	}
+	if p.PricePerSqft > 0 {
+		out = append(out, priceReason(p))
+	}
+	return compactReasons(out)
+}
+
+func compactReasons(reasons []string) []string {
+	seen := map[string]bool{}
+	out := []string{}
+	for _, r := range reasons {
+		r = strings.TrimSpace(r)
+		if r == "" || seen[r] {
+			continue
+		}
+		seen[r] = true
+		out = append(out, r)
+		if len(out) >= 4 {
+			break
+		}
+	}
+	return out
+}
+
+func priceReason(p model.Product) string {
+	if p.PricePerSqft <= 0 {
+		return ""
+	}
+	return fmt.Sprintf("价格约 $%.2f/sqft", p.PricePerSqft)
+}
+
+func productText(p model.Product) string {
+	return strings.ToLower(strings.Join([]string{
+		p.Brand, p.SeriesName, p.StyleName, p.Category, p.Material, p.ColorTone, p.Finish, p.Description,
+	}, " "))
+}
+
+func isWoodLike(p model.Product, text string) bool {
+	if p.Category == "Solid Hardwood" || p.Category == "Engineered Hardwood" || p.Category == "Laminate" || p.Category == "SPC-LVP" {
+		if materialFamilyHit(p.Material, []string{"Oak", "Maple", "Hickory", "Walnut", "Ash", "Pine", "Elm", "Teak", "Birch", "Luxury Vinyl", "Rigid Core"}) {
+			return true
+		}
+	}
+	return containsAnyFold(text, []string{"oak", "maple", "hickory", "walnut", "ash", "pine", "wood", "plank"})
+}
+
+func materialFamilyHit(material string, choices []string) bool {
+	m := strings.ToLower(material)
+	for _, choice := range choices {
+		c := strings.ToLower(choice)
+		if m == c || strings.Contains(m, c) {
+			return true
+		}
+	}
+	return false
+}
+
+func queryOptionCounts(db *sql.DB, col string, limit int) []map[string]any {
+	rows, err := db.Query("SELECT "+col+", COUNT(*) FROM products WHERE "+col+" != '' GROUP BY "+col+" ORDER BY COUNT(*) DESC, "+col+" LIMIT $1", limit)
+	if err != nil {
+		return nil
+	}
+	defer rows.Close()
+	out := []map[string]any{}
+	for rows.Next() {
+		var value string
+		var count int
+		if err := rows.Scan(&value, &count); err == nil {
+			out = append(out, map[string]any{"value": value, "count": count})
+		}
+	}
+	return out
+}
+
+func queryPriceStats(db *sql.DB) map[string]float64 {
+	rows, err := db.Query(`
+		SELECT
+			COALESCE(MIN(price_per_sqft), 0),
+			COALESCE(PERCENTILE_CONT(0.25) WITHIN GROUP (ORDER BY price_per_sqft), 0),
+			COALESCE(PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY price_per_sqft), 0),
+			COALESCE(PERCENTILE_CONT(0.75) WITHIN GROUP (ORDER BY price_per_sqft), 0),
+			COALESCE(MAX(price_per_sqft), 0)
+		FROM products WHERE price_per_sqft > 0
+	`)
+	if err != nil {
+		return map[string]float64{}
+	}
+	defer rows.Close()
+	stats := map[string]float64{}
+	if rows.Next() {
+		var min, p25, median, p75, max float64
+		if err := rows.Scan(&min, &p25, &median, &p75, &max); err == nil {
+			stats["min"] = roundOne(min)
+			stats["p25"] = roundOne(p25)
+			stats["median"] = roundOne(median)
+			stats["p75"] = roundOne(p75)
+			stats["max"] = roundOne(max)
+		}
+	}
+	return stats
+}
+
+func cleanList(values []string) []string {
+	out := []string{}
+	seen := map[string]bool{}
+	for _, v := range values {
+		v = strings.TrimSpace(v)
+		if v == "" || seen[v] {
+			continue
+		}
+		seen[v] = true
+		out = append(out, v)
+	}
+	return out
+}
+
+func toSet(values []string) map[string]bool {
+	out := map[string]bool{}
+	for _, v := range values {
+		out[v] = true
+	}
+	return out
+}
+
+func containsFold(values []string, target string) bool {
+	for _, v := range values {
+		if strings.EqualFold(v, target) {
+			return true
+		}
+	}
+	return false
+}
+
+func containsAnyFold(value string, needles []string) bool {
+	v := strings.ToLower(value)
+	for _, n := range needles {
+		if strings.Contains(v, strings.ToLower(n)) {
+			return true
+		}
+	}
+	return false
+}
+
+func clampScore(v float64) float64 {
+	if v < 0 {
+		return 0
+	}
+	if v > 100 {
+		return 100
+	}
+	return v
+}
+
+func roundOne(v float64) float64 {
+	return math.Round(v*10) / 10
+}
diff --git a/internal/repository/product_repo.go b/internal/repository/product_repo.go
index da41f44..c0f22dc 100644
--- a/internal/repository/product_repo.go
+++ b/internal/repository/product_repo.go
@@ -71,6 +71,15 @@ func QueryAllProducts(d *sql.DB) ([]model.Product, error) {
 	return prods, err
 }
 
+func QueryAllProductRows(d *sql.DB) ([]model.Product, error) {
+	rows, err := d.Query("SELECT " + selectCols + " FROM products ORDER BY id")
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+	return ScanProducts(rows)
+}
+
 func QueryFilteredProducts(d *sql.DB, p FilterParams) ([]model.Product, int, error) {
 	if p.Limit <= 0 {
 		p.Limit = 20