Update backend/agents/matcher.py

backend/agents/matcher.py

@@ -1,19 +1,21 @@
-import os
-import numpy as np
+import torch
+from transformers import AutoTokenizer, AutoModel
 from sklearn.metrics.pairwise import cosine_similarity
-from groq import Groq
 
-client = Groq(api_key=os.getenv("GROQ_API_KEY"))
+# Load once (cache in Hugging Face Space)
+tokenizer = AutoTokenizer.from_pretrained("google/embeddinggemma-300m")
+model = AutoModel.from_pretrained("google/embeddinggemma-300m")
 
 def get_embedding(text: str):
-    response = client.embeddings.create(
-        model="nomic-embed-text",  # Groq embedding model
-        input=text
-    )
-    return np.array(response.data[0].embedding)
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+    with torch.no_grad():
+        outputs = model(**inputs)
+    # Take mean pooling of last hidden state
+    embeddings = outputs.last_hidden_state.mean(dim=1)
+    return embeddings.numpy()
 
 def compute_match(resume_text: str, job_desc: str) -> float:
-    resume_emb = get_embedding(resume_text).reshape(1, -1)
-    job_emb = get_embedding(job_desc).reshape(1, -1)
+    resume_emb = get_embedding(resume_text)
+    job_emb = get_embedding(job_desc)
     score = cosine_similarity(resume_emb, job_emb)[0][0]
     return round(float(score), 3)