Update app.py

app.py

@@ -9,7 +9,7 @@ import multiprocessing
 import urllib.parse
 from datetime import datetime
 from contextlib import asynccontextmanager
-from fastapi import FastAPI, Query, Request, UploadFile, File
+from fastapi import FastAPI, Query, Request, UploadFile, File, Form
 from fastapi.responses import JSONResponse
 from fastapi.middleware.cors import CORSMiddleware
 from huggingface_hub import hf_hub_download
@@ -112,7 +112,7 @@ def load_ai_models():
         # 1. Text AI
         ai_generator = pipeline("text-generation", model="Qwen/Qwen2.5-0.5B-Instruct", device=-1, dtype=torch.float32)
         
-        # 2. Vision AI (Fix: Direct loading to bypass pipeline task errors)
+        # 2. Vision AI
         print("👁️ Loading Vision AI...")
         vision_processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
         vision_model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")
@@ -186,6 +186,7 @@ async def perform_core_search(raw_query: str):
     best_score = scored_results[0][0] if scored_results else 0
     prediction = "High confidence" if best_score > 15 else "Medium confidence" if best_score > 5 else "Low confidence"
 
+    # Yaha par hum wahi structure bana rahe hain jo frontend ko chahiye
     final_results = [{"match_score": round(s, 2), **doc} for s, doc in scored_results[:100]]
     return final_results, prediction, best_score
 
@@ -195,7 +196,6 @@ def _generate_text(messages, max_tokens: int):
     return result[0]['generated_text'][-1]['content']
 
 def _generate_image_review(image: Image):
-    # Direct model generation (Bypasses the buggy pipeline)
     inputs = vision_processor(image, return_tensors="pt")
     with torch.no_grad():
         out = vision_model.generate(**inputs, max_new_tokens=50)
@@ -219,7 +219,6 @@ async def lifespan(app: FastAPI):
     yield
 
 app = FastAPI(title="Nebulix Advanced SEO & Visual API Engine", lifespan=lifespan)
-# CORS Enabled for your separate UI
 app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"])
 
 @app.middleware("http")
@@ -241,7 +240,6 @@ async def rate_limit_middleware(request: Request, call_next):
 
     return await call_next(request)
 
-# Health Check API Endpoint (No HTML)
 @app.get("/")
 async def health_check():
     return {"status": "ok", "message": "Nebulix API Engine is running smoothly."}
@@ -267,6 +265,7 @@ async def ai_mode_short(q: str = Query(...)):
     ai_reply = await run_ai_task(messages, 150, query_key=f"short_{q}")
     return {"request_id": req_id, "ai_response": ai_reply, "results": results}
 
+# 1. PURANA CHAT API (Current UI ke liye)
 @app.get("/aimodechat")
 async def ai_mode_chat(q: str = Query(...)):
     req_id = str(uuid.uuid4())
@@ -281,6 +280,41 @@ async def ai_mode_chat(q: str = Query(...)):
     ai_reply = await run_ai_task(messages, 400, query_key=f"chat_{q}") 
     return {"request_id": req_id, "ai_response": ai_reply, "results": results}
 
+# 2. NAYA CHAT API (Future UI - Image + Text Chat ke liye)
+@app.post("/aimodechat")
+async def ai_mode_chat_vision(q: str = Form(...), file: UploadFile = File(None)):
+    req_id = str(uuid.uuid4())
+    
+    vision_context = ""
+    # Agar user ne chat me image upload ki hai, toh AI image ko decode karega
+    if file and vision_model:
+        try:
+            image_bytes = await file.read()
+            img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
+            async with ai_semaphore:
+                ai_review = await asyncio.to_thread(_generate_image_review, img)
+            vision_context = f"\n[User uploaded an image showing: {ai_review}]\n"
+        except Exception:
+            vision_context = "\n[User uploaded an image but it could not be analyzed.]\n"
+
+    # Search query combining text and vision
+    search_query = f"{q} {vision_context}".strip()
+    results, _, _ = await perform_core_search(search_query)
+    
+    context_parts = [f"Source [{i+1}]: {r.get('title')} - {r.get('description')}" for i, r in enumerate(results[:5])]
+    
+    messages = [
+        {"role": "system", "content": "You are a Smart AI Search Assistant capable of understanding text and visual context. Use the provided search sources and image details to answer accurately."},
+        {"role": "user", "content": f"User Query: {q}{vision_context}\n\nContext:\n{chr(10).join(context_parts)}"}
+    ]
+    
+    # Generate Chat response
+    cache_key = f"chat_{q}_{len(vision_context)}"
+    ai_reply = await run_ai_task(messages, 400, query_key=cache_key) 
+    
+    return {"request_id": req_id, "ai_response": ai_reply, "results": results}
+
+# IMAGE SEARCH (Fixed: Same JSON structure as normal text search)
 @app.post("/image-search")
 async def reverse_image_search(file: UploadFile = File(...)):
     if not vision_model: 
@@ -292,29 +326,29 @@ async def reverse_image_search(file: UploadFile = File(...)):
         image_bytes = await file.read()
         img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
         
+        # 1. Vision AI process karega
         async with ai_semaphore:
             ai_review = await asyncio.to_thread(_generate_image_review, img)
             
+        # 2. Text Search hoga vision ke basis pe
         raw_results, _, _ = await perform_core_search(ai_review)
         visual_results = []
         
+        # 3. DIRECTLY append the original FULL document
         for doc in raw_results:
             is_adult = str(doc.get("tag", "general")).lower() in ["adult", "nsfw"] or any(w in str(doc).lower() for w in ADULT_KEYWORDS)
             if is_adult: continue 
             
-            result_item = {
-                "title": doc.get("title", "No Title"),
-                "url": doc.get("url", ""),
-                "image_url": doc.get("image_url", doc.get("specific_data", {}).get("image", "")), 
-                "description": doc.get("description", "")
-            }
-            visual_results.append(result_item)
+            # Pura document direct bhej rahe hain bina modify kiye 
+            # jisse exact text search wala JSON format barkarar rahe
+            visual_results.append(doc)
+            
             if len(visual_results) >= 40: break
                 
+        # 4. Same structure return karenge
         return {
             "request_id": req_id,
-            "ai_review": f"Image Analysis: {ai_review}.",
-            "message": "Found visually related content and images.",
+            "query": f"Image Search Analysis: {ai_review}", # Taki UI me ye "query" ki tarah display ho
             "results_count": len(visual_results),
             "results": visual_results
         }