diff --git a/gdb_augmentor.py b/gdb_augmentor.py
new file mode 100644
index 0000000..37b0360
--- /dev/null
+++ b/gdb_augmentor.py
@@ -0,0 +1,101 @@
+import os
+import numpy as np
+from numba import njit
+import argparse
+
+ORD_A = 65
+ORD_G = 71
+
+@njit(cache=True)
+def get_popcount(n):
+    count = 0
+    while n > 0:
+        n &= (n - 1)
+        count += 1
+    return count
+
+@njit(cache=True)
+def generate_variants_core(base_arr, loc_A, loc_G, max_mutation_n_A, max_mutation_n_G):
+    n_A = len(loc_A)
+    n_G = len(loc_G)
+
+    valid_masks_A = []
+    limit_A = n_A if max_mutation_n_A < 0 else min(n_A, max_mutation_n_A)
+    for mask in range(1 << n_A):
+        if get_popcount(mask) <= limit_A:
+            valid_masks_A.append(mask)
+
+    valid_masks_G = []
+    limit_G = n_G if max_mutation_n_G < 0 else min(n_G, max_mutation_n_G)
+    for mask in range(1 << n_G):
+        if get_popcount(mask) <= limit_G:
+            valid_masks_G.append(mask)
+
+    total_variants = len(valid_masks_A) * len(valid_masks_G)
+
+    result_block = np.empty((total_variants, len(base_arr)), dtype=np.uint8)
+
+    row_idx = 0
+    for ma in valid_masks_A:
+        for mg in valid_masks_G:
+            result_block[row_idx] = base_arr
+            for bit_idx in range(n_A):
+                if (ma >> bit_idx) & 1:
+                    result_block[row_idx][loc_A[bit_idx]] = ORD_G
+            for bit_idx in range(n_G):
+                if (mg >> bit_idx) & 1:
+                    result_block[row_idx][loc_G[bit_idx]] = ORD_A
+            row_idx += 1
+            
+    return result_block
+
+def run_augmentation(input_file, output_file, max_mutation_n_A=None, max_mutation_n_G=None):
+    with open(input_file, 'r') as f:
+        lines = [line.strip() for line in f.readlines() if line.strip()]
+
+    if lines:
+        header = lines[0]
+        data_lines = lines[1:]
+    else:
+        return
+
+    unique_structs = set()
+    limit_A = -1 if max_mutation_n_A is None else max_mutation_n_A
+    limit_G = -1 if max_mutation_n_G is None else max_mutation_n_G
+
+    for line in data_lines:
+        base_arr = np.frombuffer(line.encode('ascii'), dtype=np.uint8)
+        loc_A = np.where(base_arr == ORD_A)[0]
+        loc_G = np.where(base_arr == ORD_G)[0]
+
+        variants_block = generate_variants_core(base_arr, loc_A, loc_G, limit_A, limit_G)
+
+        for row in variants_block:
+            unique_structs.add(row.tobytes().decode('ascii'))
+            
+    sorted_result = sorted(list(unique_structs), key=lambda x: (len(x), x))
+
+    with open(output_file, 'w') as f:
+        f.write(header + "\n")
+        f.write("\n".join(sorted_result) + "\n")
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        prog="gdb_augmentor",
+        description="Augment glycan structures in GDB file by swapping A/G residues.",
+    )
+    
+    parser.add_argument("-i", "--input_gdb", type=str, required=True, help="Path to input .gdb file")
+    parser.add_argument("-o", "--output_gdb", type=str, required=True, help="Path to output .gdb file")
+
+    parser.add_argument("-ma", "--max_mutation_a", type=int, default=2, help="Max number of 'A' to swap (default: 2)")
+    parser.add_argument("-mg", "--max_mutation_g", type=int, default=2, help="Max number of 'G' to swap (default: 2)")
+    
+    args = parser.parse_args()
+    
+    run_augmentation(
+        input_file=args.input_gdb,
+        output_file=args.output_gdb,
+        max_mutation_n_A=args.max_mutation_a,
+        max_mutation_n_G=args.max_mutation_g,
+    )
\ No newline at end of file