diff --git a/src/data.py b/src/data.py
index c902c94..288caf2 100644
--- a/src/data.py
+++ b/src/data.py
@@ -28,7 +28,7 @@ def preprocess_tweet(tweet: str) -> str:
     # Unescape HTML characters
     tweet = html.unescape(tweet)
 
-    # Replace URLs wiht URL token
+    # Replace URLs with URL token
     tweet = re.sub(r"http\S+", "[URL]", tweet)
 
     # Replace @mentions with @USER token
diff --git a/src/mixtral_tweets.py b/src/mixtral_tweets.py
new file mode 100644
index 0000000..fed7d5a
--- /dev/null
+++ b/src/mixtral_tweets.py
@@ -0,0 +1,67 @@
+import requests
+import tqdm
+import csv
+import os
+
+# Third party
+from dotenv import load_dotenv
+from datasets import load_dataset
+
+
+def get_api_response(sampled_tweets: list):
+    prompt_start = (
+        "Create synthetic BEARISH tweets about the financial market. Examples:"
+    )
+    sampled_tweets = sampled_tweets.join("\n")
+
+    return requests.post(
+        "https://api.together.xyz/inference",
+        json={
+            "model": "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
+            "max_tokens": 512,
+            "prompt": "",
+            "request_type": "language-model-inference",
+            "temperature": 1.05,
+            "top_p": 1,
+            "top_k": 100,
+            "repetition_penalty": 1.3,
+            "stop": ["<|im_end|>", "<|im_start|>"],
+            "messages": [
+                {
+                    "content": f"{prompt_start}\n{sampled_tweets}",
+                    "role": "user",
+                },
+            ],
+            "prompt_format_string": "<|im_start|>user\n {prompt}\n<|im_end|>\n<|im_start|>assistant\n",
+            "repetitive_penalty": 1.3,
+        },
+        headers={
+            "Authorization": f"Bearer {key}",
+        },
+    )
+
+
+if __name__ == "__main__":
+    load_dotenv()
+    key = os.getenv("TOGETHER_API")
+
+    dataset = load_dataset(
+        "TimKoornstra/financial-tweets-sentiment",
+        split="train",
+        cache_dir="data/finetune/",
+        # download_mode="force_redownload",
+    )
+    dataframe = dataset.to_pandas()
+
+    # Open the CSV file. The 'a' mode appends to the file without truncating it.
+    with open("bearish-tweets.csv", "a", newline="", encoding="utf-8") as f:
+        writer = csv.writer(f)
+
+        for i in tqdm.tqdm(range(1000)):
+            # Random sample of 10 tweets, where the label = 2 (bearish)
+            df_sample = dataframe[dataframe["sentiment"] == 2].sample(10)
+            res = get_api_response(df_sample["tweet"].tolist())
+            generated = res.json()["output"]["choices"][0]["text"].split("\n")
+
+            for tweet in generated:
+                writer.writerow([tweet])