diff --git a/backend/src/landmarks_manager.py b/backend/src/landmarks_manager.py
index 37eb032..618e169 100644
--- a/backend/src/landmarks_manager.py
+++ b/backend/src/landmarks_manager.py
@@ -1,8 +1,8 @@
 import math as m
 import json, os
 
-from typing import List, Tuple
-from OSMPythonTools.overpass import Overpass, overpassQueryBuilder, Nominatim
+from typing import List, Tuple, Optional
+from OSMPythonTools.overpass import Overpass, overpassQueryBuilder
 
 from structs.landmarks import Landmark, LandmarkType
 from structs.preferences import Preferences, Preference
@@ -38,7 +38,9 @@ def generate_landmarks(preferences: Preferences, coordinates: Tuple[float, float
         correct_score(L3, preferences.shopping)
         L += L3
 
-    return remove_duplicates(L), take_most_important(L)
+    L = remove_duplicates(L)
+
+    return L, take_most_important(L)
 
 
 """def generate_landmarks(preferences: Preferences, city_country: str = None, coordinates: Tuple[float, float] = None) -> Tuple[List[Landmark], List[Landmark]] :
@@ -91,7 +93,7 @@ def get_list(path: str) -> List[str] :
 
 
 # Take the most important landmarks from the list
-def take_most_important(L: List[Landmark]) -> List[Landmark] :
+def take_most_important(L: List[Landmark], N = 0) -> List[Landmark] :
     
     # Read the parameters from the file
     with open (os.path.dirname(os.path.abspath(__file__)) + '/parameters/landmarks_manager.params', "r") as f :
@@ -125,7 +127,7 @@ def take_most_important(L: List[Landmark]) -> List[Landmark] :
     for i, elem in enumerate(L_copy) :
         scores[i] = elem.attractiveness
 
-    res = sorted(range(len(scores)), key = lambda sub: scores[sub])[-N_important:]
+    res = sorted(range(len(scores)), key = lambda sub: scores[sub])[-(N_important-N):]
 
     for i, elem in enumerate(L_copy) :
         if i in res :
@@ -148,21 +150,15 @@ def remove_duplicates(L: List[Landmark]) -> List[Landmark] :
 
     L_clean = []
     names = []
-    coords = []
 
     for landmark in L :
-        if landmark.name in names and landmark.location in coords: 
+        if landmark.name in names: 
             continue  
-        
-        approx_coords = tuple((round(landmark.location[0], 4), round(landmark.location[0], 4)))
-        
-        if approx_coords in coords :
-            continue   
+
 
         else :
             names.append(landmark.name)
             L_clean.append(landmark)
-            coords.append(approx_coords)
     
     return L_clean
     
diff --git a/backend/src/optimizer.py b/backend/src/optimizer.py
index 76fd0ed..d45d51b 100644
--- a/backend/src/optimizer.py
+++ b/backend/src/optimizer.py
@@ -4,12 +4,13 @@ import json, os
 from typing import List, Tuple
 from scipy.optimize import linprog
 from math import radians, sin, cos, acos
+from shapely import Polygon
 
 from structs.landmarks import Landmark
 
     
 # Function to print the result
-def print_res(L: List[Landmark], L_tot) -> list:
+def print_res(L: List[Landmark], L_tot):
 
     if len(L) == L_tot: 
         print('\nAll landmarks can be visited within max_steps, the following order is suggested : ')
@@ -25,10 +26,10 @@ def print_res(L: List[Landmark], L_tot) -> list:
             print('- ' + elem.name)
 
     print("\nMinutes walked : " + str(dist))
-    print(f"Visited {len(L)} out of {L_tot} landmarks")
+    print(f"Visited {len(L)-2} out of {L_tot-2} landmarks")
 
 
-# Prevent the use of a particular set of nodes
+# Prevent the use of a particular solution
 def prevent_config(resx, A_ub, b_ub) -> bool:
     
     for i, elem in enumerate(resx):
@@ -56,7 +57,7 @@ def prevent_config(resx, A_ub, b_ub) -> bool:
     return A_ub, b_ub
 
 
-# Prevent the possibility of a given set of vertices
+# Prevent the possibility of a given solution bit
 def break_cricle(circle_vertices: list, L: int, A_ub: list, b_ub: list) -> bool:
 
     if L-1 in circle_vertices :
@@ -73,7 +74,7 @@ def break_cricle(circle_vertices: list, L: int, A_ub: list, b_ub: list) -> bool:
     return A_ub, b_ub
 
 
-# Checks if the path is connected, returns a circle if it finds one
+# Checks if the path is connected, returns a circle if it finds one and the RESULT
 def is_connected(resx) -> bool:
     
     # first round the results to have only 0-1 values
@@ -182,8 +183,8 @@ def init_ub_dist(landmarks: List[Landmark], max_steps: int):
     return c, A_ub, [max_steps]
 
 
-# Constraint to respect max number of travels
-def respect_number(L, A_ub, b_ub):
+# Constraint to respect only one travel per landmark. Also caps the total number of visited landmarks
+def respect_number(L:int, A_ub, b_ub):
 
     ones = [1]*L
     zeros = [0]*L
@@ -192,6 +193,14 @@ def respect_number(L, A_ub, b_ub):
         A_ub = np.vstack((A_ub, h))
         b_ub.append(1)
 
+    # Read the parameters from the file
+    with open (os.path.dirname(os.path.abspath(__file__)) + '/parameters/optimizer.params', "r") as f :
+        parameters = json.loads(f.read())
+        max_landmarks = parameters['max landmarks']
+
+    A_ub = np.vstack((A_ub, ones*L))
+    b_ub.append(max_landmarks+1)
+
     return A_ub, b_ub
 
 
@@ -290,7 +299,7 @@ def respect_order(N: int, A_eq, b_eq):
     return A_eq, b_eq
 
 
-# Computes the path length given path matrix (dist_table) and a result
+# Computes the time to reach from each landmark to the next
 def add_time_to_reach(order: List[int], landmarks: List[Landmark])->List[Landmark] :
     
     # Read the parameters from the file
@@ -315,6 +324,7 @@ def add_time_to_reach(order: List[int], landmarks: List[Landmark])->List[Landmar
     
     return L
 
+
 def add_time_to_reach_simple(ordered_visit: List[Landmark])-> List[Landmark] :
     
     # Read the parameters from the file
@@ -326,14 +336,17 @@ def add_time_to_reach_simple(ordered_visit: List[Landmark])-> List[Landmark] :
     L =  []
     prev = ordered_visit[0]
     L.append(prev)
+    total_dist = 0
 
     for elem in ordered_visit[1:] :
         elem.time_to_reach = get_distance(elem.location, prev.location, detour_factor, speed)[1]
         elem.must_do = True
         L.append(elem)
         prev = elem
+        total_dist += get_distance(elem.location, prev.location, detour_factor, speed)[1]
+        
 
-    return L
+    return L, total_dist
 
 
 # Main optimization pipeline
@@ -366,22 +379,23 @@ def solve_optimization (landmarks :List[Landmark], max_steps: int, printing_deta
     else :
         order, circle = is_connected(res.x)
         i = 0
-        timeout = 300
+        timeout = 80
         while len(circle) != 0 and i < timeout:
             A_ub, b_ub = prevent_config(res.x, A_ub, b_ub)
             A_ub, b_ub = break_cricle(order, len(landmarks), A_ub, b_ub)
             res = linprog(c, A_ub=A_ub, b_ub=b_ub, A_eq=A_eq, b_eq = b_eq, bounds=x_bounds, method='highs', integrality=3)
             order, circle = is_connected(res.x)
             if len(circle) == 0 :
-                # Add the times to reach and stop optimizing
-                L = add_time_to_reach(order, landmarks)
                 break
-            #print(i)
+            print(i)
             i += 1
         
         if i == timeout :
             raise TimeoutError(f"Optimization took too long. No solution found after {timeout} iterations.")
 
+        # Add the times to reach and stop optimizing
+        L = add_time_to_reach(order, landmarks)
+            
         if printing_details is True :
             if i != 0 :
                 print(f"Neded to recompute paths {i} times because of unconnected loops...")
diff --git a/backend/src/parameters/landmarks_manager.params b/backend/src/parameters/landmarks_manager.params
index 087525e..942c214 100644
--- a/backend/src/parameters/landmarks_manager.params
+++ b/backend/src/parameters/landmarks_manager.params
@@ -4,5 +4,5 @@
   "church coeff" : 0.6,
   "park coeff" : 1.5,
   "tag coeff" : 100,
-  "N important" : 30
+  "N important" : 40
 }
\ No newline at end of file
diff --git a/backend/src/parameters/optimizer.params b/backend/src/parameters/optimizer.params
index 5d32077..18ca240 100644
--- a/backend/src/parameters/optimizer.params
+++ b/backend/src/parameters/optimizer.params
@@ -1,4 +1,5 @@
 {
   "detour factor" : 1.4,
-  "average walking speed" : 4.8
+  "average walking speed" : 4.8,
+  "max landmarks" : 10
 }
\ No newline at end of file
diff --git a/backend/src/refiner.py b/backend/src/refiner.py
index b57610c..dec6101 100644
--- a/backend/src/refiner.py
+++ b/backend/src/refiner.py
@@ -1,10 +1,15 @@
+from collections import defaultdict
+from heapq import heappop, heappush
+from itertools import permutations
+import os, json
+
 from shapely import buffer, LineString, Point, Polygon, MultiPoint, convex_hull, concave_hull, LinearRing
-from typing import List
+from typing import List, Tuple
 from math import pi
 
 from structs.landmarks import Landmark
 from landmarks_manager import take_most_important
-from optimizer import solve_optimization, add_time_to_reach_simple, print_res
+from optimizer import solve_optimization, add_time_to_reach_simple, print_res, get_distance
 
 
 def create_corridor(landmarks: List[Landmark], width: float) :
@@ -28,10 +33,135 @@ def create_linestring(landmarks: List[Landmark])->List[Point] :
 
 
 def is_in_area(area: Polygon, coordinates) -> bool :
-
   point = Point(coordinates)
   return point.within(area)
 
+
+def is_close_to(location1: Tuple[float], location2: Tuple[float]):
+    """Determine if two locations are close by rounding their coordinates to 3 decimals."""
+    absx = abs(location1[0] - location2[0])
+    absy = abs(location1[1] - location2[1])
+
+    return absx < 0.001 and absy < 0.001
+    #return (round(location1[0], 3), round(location1[1], 3)) == (round(location2[0], 3), round(location2[1], 3))
+
+
+def rearrange(landmarks: List[Landmark]) -> List[Landmark]:
+  
+  i = 1
+  while i < len(landmarks):
+    j = i+1
+    while j < len(landmarks):
+      if is_close_to(landmarks[i].location, landmarks[j].location) and landmarks[i].name not in ['start', 'finish'] and landmarks[j].name not in ['start', 'finish']:
+        # If they are not adjacent, move the j-th element to be adjacent to the i-th element
+        if j != i + 1:
+          landmarks.insert(i + 1, landmarks.pop(j))
+        break  # Move to the next i-th element after rearrangement
+      j += 1
+    i += 1
+  
+  return landmarks
+
+"""
+def find_shortest_path(landmarks: List[Landmark]) -> List[Landmark]:
+    
+  # Read from data
+  with open (os.path.dirname(os.path.abspath(__file__)) + '/parameters/optimizer.params', "r") as f :
+    parameters = json.loads(f.read())
+    detour = parameters['detour factor']
+    speed = parameters['average walking speed']
+
+  # Step 1: Build the graph
+  graph = defaultdict(list)
+  for i in range(len(landmarks)):
+    for j in range(len(landmarks)):
+      if i != j:
+        distance = get_distance(landmarks[i].location, landmarks[j].location, detour, speed)[1]
+        graph[i].append((distance, j))
+
+  # Step 2: Dijkstra's algorithm to find the shortest path from start to finish
+  start_idx = next(i for i, lm in enumerate(landmarks) if lm.name == 'start')
+  finish_idx = next(i for i, lm in enumerate(landmarks) if lm.name == 'finish')
+
+  distances = {i: float('inf') for i in range(len(landmarks))}
+  previous_nodes = {i: None for i in range(len(landmarks))}
+  distances[start_idx] = 0
+  priority_queue = [(0, start_idx)]
+
+  while priority_queue:
+    current_distance, current_index = heappop(priority_queue)
+
+    if current_distance > distances[current_index]:
+      continue
+
+    for neighbor_distance, neighbor_index in graph[current_index]:
+      distance = current_distance + neighbor_distance
+
+      if distance < distances[neighbor_index]:
+        distances[neighbor_index] = distance
+        previous_nodes[neighbor_index] = current_index
+        heappush(priority_queue, (distance, neighbor_index))
+
+  # Step 3: Backtrack from finish to start to find the path
+  path = []
+  current_index = finish_idx
+  while current_index is not None:
+    path.append(landmarks[current_index])
+    current_index = previous_nodes[current_index]
+  path.reverse()
+
+  return path
+"""
+"""
+def total_path_distance(path: List[Landmark], detour, speed) -> float:
+  total_distance = 0
+  for i in range(len(path) - 1):
+    total_distance += get_distance(path[i].location, path[i + 1].location, detour, speed)[1]
+  return total_distance
+"""
+
+
+def find_shortest_path_through_all_landmarks(landmarks: List[Landmark]) -> List[Landmark]:
+    
+  # Read from data
+  with open (os.path.dirname(os.path.abspath(__file__)) + '/parameters/optimizer.params', "r") as f :
+    parameters = json.loads(f.read())
+    detour = parameters['detour factor']
+    speed = parameters['average walking speed']
+    
+  # Step 1: Find 'start' and 'finish' landmarks
+  start_idx = next(i for i, lm in enumerate(landmarks) if lm.name == 'start')
+  finish_idx = next(i for i, lm in enumerate(landmarks) if lm.name == 'finish')
+  
+  start_landmark = landmarks[start_idx]
+  finish_landmark = landmarks[finish_idx]
+  
+
+  # Step 2: Create a list of unvisited landmarks excluding 'start' and 'finish'
+  unvisited_landmarks = [lm for i, lm in enumerate(landmarks) if i not in [start_idx, finish_idx]]
+  
+  # Step 3: Initialize the path with the 'start' landmark
+  path = [start_landmark]
+  coordinates = [landmarks[start_idx].location]
+
+  current_landmark = start_landmark
+  
+  # Step 4: Use nearest neighbor heuristic to visit all landmarks
+  while unvisited_landmarks:
+    nearest_landmark = min(unvisited_landmarks, key=lambda lm: get_distance(current_landmark.location, lm.location, detour, speed)[1])
+    path.append(nearest_landmark)
+    coordinates.append(nearest_landmark.location)
+    current_landmark = nearest_landmark
+    unvisited_landmarks.remove(nearest_landmark)
+
+  # Step 5: Finally add the 'finish' landmark to the path
+  path.append(finish_landmark)
+  coordinates.append(landmarks[finish_idx].location)
+
+  path_poly = Polygon(coordinates)
+
+  return path, path_poly
+
 def get_minor_landmarks(all_landmarks: List[Landmark], visited_landmarks: List[Landmark], width: float) -> List[Landmark] :
 
   second_order_landmarks = []
@@ -45,7 +175,7 @@ def get_minor_landmarks(all_landmarks: List[Landmark], visited_landmarks: List[L
     if is_in_area(area, landmark.location) and landmark.name not in visited_names:
       second_order_landmarks.append(landmark)
 
-  return take_most_important(second_order_landmarks)
+  return take_most_important(second_order_landmarks, len(visited_landmarks))
 
 
 
@@ -58,55 +188,106 @@ def get_minor_landmarks(all_landmarks: List[Landmark], visited_landmarks: List[L
   full_set = base_tour[:-1] + minor_landmarks   # create full set of possible landmarks (without finish)
   full_set.append(base_tour[-1])                # add finish back
 
-  new_route = solve_optimization(full_set, max_time, print_infos)
+  new_tour = solve_optimization(full_set, max_time, print_infos)
 
-  return new_route"""
+  return new_tour"""
 
 
 def refine_optimization(landmarks: List[Landmark], base_tour: List[Landmark], max_time: int, print_infos: bool) -> List[Landmark] :
 
+  # Read from the file
+  with open (os.path.dirname(os.path.abspath(__file__)) + '/parameters/optimizer.params', "r") as f :
+    parameters = json.loads(f.read())
+    max_landmarks = parameters['max landmarks']
+  
+  if len(base_tour)-2 >= max_landmarks :
+    return base_tour
+  
+
   minor_landmarks = get_minor_landmarks(landmarks, base_tour, 200)
 
-  if print_infos : print("There are " + str(len(minor_landmarks)) + " minor landmarks around the predicted path")
+  if print_infos : print("Using " + str(len(minor_landmarks)) + " minor landmarks around the predicted path")
 
   # full set of visitable landmarks
   full_set = base_tour[:-1] + minor_landmarks   # create full set of possible landmarks (without finish)
   full_set.append(base_tour[-1])                # add finish back
 
-  # get a new route
-  new_route = solve_optimization(full_set, max_time, False)
-
-  coords = []         # Coordinates of the new route
-  coords_dict = {}    # maps the location of an element to the element itself. Used to access the elements back once we get the geometry
-
-  # Iterate through the new route without finish
-  for elem in new_route[:-1] :
-    coords.append(Point(elem.location))
-    coords_dict[elem.location] = elem         # if start = goal, only finish remains
-
-  # Create a concave polygon using the coordinates
-  better_route_poly = concave_hull(MultiPoint(coords))  # Create concave hull with "core" of route leaving out start and finish
-  xs, ys = better_route_poly.exterior.xy
-
-  better_route = []   # List of ordered visit
-  name_index = {}     # Maps the name of a landmark to its index in the concave polygon
-
-  # Loop through the polygon and generate the better (ordered) route
-  for i,x in enumerate(xs[:-1]) :
-    better_route.append(coords_dict[tuple((x,ys[i]))])
-    name_index[coords_dict[tuple((x,ys[i]))].name] = i
-
+  # get a new tour
+  new_tour = solve_optimization(full_set, max_time, False)
+  new_tour, new_dist = add_time_to_reach_simple(new_tour)
   
-  # Scroll the list to have start in front again
-  start_index = name_index['start']
-  better_route = better_route[start_index:] + better_route[:start_index]
+  """#if base_tour[0].location == base_tour[-1].location :
+  if False :
+    coords = []         # Coordinates of the new tour
+    coords_dict = {}    # maps the location of an element to the element itself. Used to access the elements back once we get the geometry
 
-  # Append the finish back and correct the time to reach
-  better_route.append(new_route[-1])
-  better_route = add_time_to_reach_simple(better_route)
+    # Iterate through the new tour without finish
+    for elem in new_tour[:-1] :
+      coords.append(Point(elem.location))
+      coords_dict[elem.location] = elem         # if start = goal, only finish remains
+
+    # Create a concave polygon using the coordinates
+    better_tour_poly = concave_hull(MultiPoint(coords))  # Create concave hull with "core" of tour leaving out start and finish
+    xs, ys = better_tour_poly.exterior.xy
+
+    # reverse the xs and ys
+    xs.reverse()
+    ys.reverse()
+
+    better_tour = []   # List of ordered visit
+    name_index = {}     # Maps the name of a landmark to its index in the concave polygon
+
+    # Loop through the polygon and generate the better (ordered) tour
+    for i,x in enumerate(xs[:-1]) :
+      better_tour.append(coords_dict[tuple((x,ys[i]))])
+      name_index[coords_dict[tuple((x,ys[i]))].name] = i
+
+
+    # Scroll the list to have start in front again
+    start_index = name_index['start']
+    better_tour = better_tour[start_index:] + better_tour[:start_index]
+
+    # Append the finish back and correct the time to reach
+    better_tour.append(new_tour[-1])
+
+    # Rearrange only if polygon
+    better_tour = rearrange(better_tour)
+
+    # Add the time to reach
+    better_tour = add_time_to_reach_simple(better_tour)
+  """
+  
+
+  """
+  if not better_poly.is_simple :
+    
+    coords_dict = {}
+    better_tour2 = []
+    for elem in better_tour :
+      coords_dict[elem.location] = elem
+
+    better_poly2 = better_poly.buffer(0)
+    new_coords = better_poly2.exterior.coords[:]
+    start_coords = base_tour[0].location
+    start_index = new_coords.
+
+    #for point in new_coords :
+  """
+
+
+  better_tour, better_poly = find_shortest_path_through_all_landmarks(new_tour)
+  better_tour, better_dist = add_time_to_reach_simple(better_tour)
+
+  if new_dist < better_dist :
+    final_tour = new_tour
+  else :
+    final_tour = better_tour
 
   if print_infos :
+    print("\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n")
     print("\nRefined tour (result of second stage optimization): ")
-    print_res(better_route, len(better_route))
+    print_res(final_tour, len(full_set))
 
-  return better_route
+
+  
+  return final_tour
diff --git a/backend/src/tester.py b/backend/src/tester.py
index 4920847..758e003 100644
--- a/backend/src/tester.py
+++ b/backend/src/tester.py
@@ -84,6 +84,10 @@ def test4(coordinates: tuple[float, float]) -> List[Landmark]:
     # Create start and finish 
     start = Landmark(name='start', type=LandmarkType(landmark_type='start'), location=coordinates, osm_type='start', osm_id=0, attractiveness=0, must_do=True, n_tags = 0)
     finish = Landmark(name='finish', type=LandmarkType(landmark_type='finish'), location=coordinates, osm_type='finish', osm_id=0, attractiveness=0, must_do=True, n_tags = 0)
+    #finish = Landmark(name='finish', type=LandmarkType(landmark_type='finish'), location=(48.8777055, 2.3640967), osm_type='finish', osm_id=0, attractiveness=0, must_do=True, n_tags = 0)
+    #start = Landmark(name='start', type=LandmarkType(landmark_type='start'), location=(48.847132, 2.312359), osm_type='start', osm_id=0, attractiveness=0, must_do=True, n_tags = 0)
+    #finish = Landmark(name='finish', type=LandmarkType(landmark_type='finish'), location=(48.843185, 2.344533), osm_type='finish', osm_id=0, attractiveness=0, must_do=True, n_tags = 0)
+    #finish = Landmark(name='finish', type=LandmarkType(landmark_type='finish'), location=(48.847132, 2.312359), osm_type='finish', osm_id=0, attractiveness=0, must_do=True, n_tags = 0)
     
     # Generate the landmarks from the start location
     landmarks, landmarks_short = generate_landmarks(preferences=preferences, coordinates=start.location)
@@ -94,7 +98,7 @@ def test4(coordinates: tuple[float, float]) -> List[Landmark]:
     landmarks_short.append(finish)
 
     # TODO use these parameters in another way
-    max_walking_time = 4    # hours
+    max_walking_time = 2    # hours
     detour = 30             # minutes
 
     # First stage optimization
@@ -107,5 +111,6 @@ def test4(coordinates: tuple[float, float]) -> List[Landmark]:
 
 
 #test4(tuple((48.8344400, 2.3220540)))       # Café Chez César 
-test4(tuple((48.8375946, 2.2949904)))       # Point random
+#test4(tuple((48.8375946, 2.2949904)))       # Point random
+test4(tuple((47.377859, 8.540585)))         # Zurich HB
 #test3('Vienna, Austria')
\ No newline at end of file