diff --git a/nlu/models/slot_filler.py b/nlu/models/slot_filler.py
index d4d3182..11c509e 100644
--- a/nlu/models/slot_filler.py
+++ b/nlu/models/slot_filler.py
@@ -1,3 +1,4 @@
+import os
 import functools
 import json
 from transformers import TFBertModel, AutoTokenizer
@@ -29,12 +30,6 @@ import boilerplate as tfbp
 # the paper with the original dataset.
 ##
 
-BERT_MODEL_BY_LANGUAGE = {
-    'en': "bert-base-cased",
-    'fr': "dbmdz/bert-base-french-europeana-cased",
-}
-
-
 @tfbp.default_export
 class SlotFiller(tfbp.Model):
     default_hparams = {
@@ -53,7 +48,24 @@ class SlotFiller(tfbp.Model):
 
         # Load Tokenizer from transformers
         # We will use a pretrained bert model bert-base-cased for both Tokenizer and our classifier.
-        bert_model_name = BERT_MODEL_BY_LANGUAGE[self.hparams.language or "en"]
+        
+        # Read the environment variable
+        bert_model_by_language_json = os.getenv('BERT_MODEL_BY_LANGUAGE_JSON')
+
+        # Check if the environment variable is set
+        if not bert_model_by_language_json:
+            raise ValueError("The BERT_MODEL_BY_LANGUAGE_JSON environment variable is not set.")
+
+        # Parse the JSON string into a Python dictionary
+        try:
+            bert_models = json.loads(bert_model_by_language_json)
+        except json.JSONDecodeError as e:
+            raise ValueError(f"Failed to parse BERT_MODEL_BY_LANGUAGE_JSON: {e}")
+        
+        try:
+            bert_model_name = bert_models[self.hparams.language or "en"]
+        except json.JSONDecodeError as e:
+            raise ValueError(f"No Bert model is available for the provided language: {e}")
 
         self.tokenizer = AutoTokenizer.from_pretrained(
             bert_model_name, use_fast=False)