update

api/main.py

@@ -1,6 +1,7 @@
 # api/main.py
 
-from fastapi import FastAPI
+from fastapi import FastAPI, HTTPException
+
 from fastapi.middleware.cors import CORSMiddleware
 
 from api.schemas import AnalyzeRequest, ChatRequest, ChatResponse
@@ -46,16 +47,22 @@ def health():
 
 @app.post("/analyze")
 def analyze(data: AnalyzeRequest):
-
-    result = analyze_project(
-        title=data.title,
-        description=data.description,
-        abstract=data.abstract,
-        features=data.features,
-        top_k=data.top_k
-    )
-
-    return result
+    try:
+        result = analyze_project(
+            title=data.title,
+            description=data.description,
+            abstract=data.abstract,
+            features=data.features,
+            top_k=data.top_k
+        )
+        return result
+    except HTTPException:
+        raise
+    except Exception as exc:
+        raise HTTPException(
+            status_code=500,
+            detail=f"Analysis failed: {exc}"
+        )
 
 
 @app.post("/chat", response_model=ChatResponse)

api/services.py

@@ -199,7 +199,8 @@ def analyze_project(
     if not isinstance(results, pd.DataFrame) or len(results) == 0:
         return {
             "message": "No similar projects found",
-            "extracted_features": merged
+            "extracted_features": merged,
+            "overall_originality_score": 100.0
         }
 
     # -----------------------------------
@@ -216,12 +217,17 @@ def analyze_project(
             "final_originality_score": round(float(row.get("originality_score", 0)), 4)
         })
 
+    # Overall = worst-case originality (against the most similar project)
+    overall_originality_score = top_projects[0]["final_originality_score"]
+
     return {
         "extracted_features": merged,
+        "overall_originality_score": overall_originality_score,
         "top_similar_projects": top_projects
     }
 
 
+
 def chat_with_llm(user_id: str, message: str):
     try:
         from src.recommendation_engine.chatbot_engine import chatbot

src/similarity_model/hybrid_ranker.py

@@ -53,24 +53,20 @@ def get_dynamic_weights(
     coverage: float
 ):
     """
-    Adaptive weights depending on
-    feature richness.
+    Adaptive weights depending on feature richness.
+    Returns (semantic_w, feature_w, coverage_w) — always sum to 1.0
     """
 
-    semantic_w = DEFAULT_SEMANTIC_WEIGHT
-    feature_w = DEFAULT_FEATURE_WEIGHT
-
-    # many strong features
+    # Rich features → trust feature matching more
     if feature_count >= 5 and coverage >= 0.60:
-        semantic_w = 0.30
-        feature_w = HIGH_FEATURE_WEIGHT
+        return 0.40, 0.45, 0.15
 
-    # weak features -> trust semantic more
-    elif feature_count <= 2:
-        semantic_w = 0.80
-        feature_w = LOW_FEATURE_WEIGHT
+    # Sparse features → trust semantic more
+    if feature_count <= 2:
+        return 0.70, 0.20, 0.10
 
-    return semantic_w, feature_w
+    # Default balance
+    return 0.55, 0.35, 0.10
 
 
 # =====================================================
@@ -85,10 +81,19 @@ def compute_hybrid_score(
 ) -> float:
 
     semantic_score = clamp(semantic_score)
-    feature_score = clamp(feature_score)
-    coverage = clamp(coverage)
+    feature_score  = clamp(feature_score)
+    coverage       = clamp(coverage)
+
+    shared_count = feature_count - unique_query_count
+
+    # ==========================================
+    # Near-duplicate fast path
+    # ==========================================
+    if shared_count >= 6 and unique_query_count <= 1:
+        return 0.95
+
     # ==========================================
-    # Strong feature overlap case
+    # Strong feature overlap fast path
     # ==========================================
     if coverage >= 0.90 and feature_score >= 0.65:
         return round(
@@ -101,31 +106,23 @@ def compute_hybrid_score(
         )
 
     # ==========================================
-    # Normal scoring
+    # No feature overlap → rely on semantics only
     # ==========================================
-    shared_ratio = (
-        (feature_count - unique_query_count)
-        / max(feature_count, 1)
-    )
-
-    score = (
-        0.90 * (shared_ratio ** 2.0)
-        + 0.07 * feature_score
-        + 0.03 * semantic_score
-    )
-
-    # No feature overlap
-    # No feature overlap
     if feature_score == 0 or coverage == 0:
-        return 0.03
+        return round(clamp(0.20 * semantic_score), 4)
 
-    shared_count = (
-        feature_count - unique_query_count
+    # ==========================================
+    # Normal scoring with dynamic weights
+    # ==========================================
+    semantic_w, feature_w, coverage_w = get_dynamic_weights(
+        feature_count, coverage
     )
 
-    # Near duplicate    
-    if shared_count >= 6 and unique_query_count <= 1:
-        return 0.95
+    score = (
+        semantic_w * semantic_score +
+        feature_w  * feature_score +
+        coverage_w * coverage
+    )
 
     return round(clamp(score), 4)
 

src/similarity_model/preprocessing.py

@@ -4,6 +4,7 @@
 
 import re
 import logging
+from functools import lru_cache
 from pathlib import Path
 import pandas as pd
 from sentence_transformers import SentenceTransformer
@@ -13,6 +14,7 @@ from .llm_feature_extractor import (
 )
 
 
+
 # =====================================================
 # Logging
 # =====================================================
@@ -27,9 +29,12 @@ logger = logging.getLogger(__name__)
 # =====================================================
 MODEL_NAME = "all-MiniLM-L6-v2"
 
-embed_model = SentenceTransformer(
-    MODEL_NAME
-)
+@lru_cache(maxsize=1)
+def _get_embed_model():
+    """Lazy-load the embedding model once on first use."""
+    logger.info(f"Loading embed model: {MODEL_NAME}")
+    return SentenceTransformer(MODEL_NAME)
+
 
 # =====================================================
 # Config
@@ -119,7 +124,7 @@ def semantic_deduplicate(
 # =====================================================
 def extract_features(text):
 
-    print("USING GEMINI FEATURE EXTRACTOR")
+    logger.info("Using Gemini feature extractor")
 
     features = extract_features_llm(
         text
@@ -127,11 +132,12 @@ def extract_features(text):
 
     return semantic_deduplicate(
         features,
-        embed_model,
+        _get_embed_model(),
         threshold=0.85
     )
 
 
+
 # =====================================================
 # Main Pipeline
 # =====================================================

src/similarity_model/similarity_engine.py

@@ -39,10 +39,11 @@ TITLE_COL = "project_title"
 FEATURE_COL = "features"
 
 DEFAULT_TOP_K = 5
-DEFAULT_SEARCH_POOL = 20
+DEFAULT_SEARCH_POOL = 50
 DEFAULT_MIN_SEMANTIC_SCORE = 0.30
 MAX_QUERY_FEATURES = 12
 
+
 # =====================================================
 # Query Builders
 # =====================================================
@@ -186,11 +187,13 @@ def compare_candidate(
 
     hybrid_score = base_similarity
 
-    originality_score = round(
-        (1.0 - hybrid_score) * 100,
-        2
+    originality_score = compute_originality(
+        hybrid_score=hybrid_score,
+        unique_query_features=unique_query_count,
+        total_query_features=query_feature_count
     )
 
+
     confidence_score = compute_confidence(
         semantic_score=semantic_score,
         feature_score=feature_score,