Update backend.py

backend.py

@@ -7,7 +7,7 @@ from flask import Flask, request, jsonify
 from flask_cors import CORS
 from dotenv import load_dotenv
 from faster_whisper import WhisperModel
-import anthropic
+import google.generativeai as genai
 from supabase import create_client
 
 load_dotenv()
@@ -21,7 +21,7 @@ print("Loading Whisper model (base)…")
 whisper_model = WhisperModel("base", device="cpu", compute_type="int8")
 print("Whisper ready.")
 
-claude = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
+genai.configure(api_key=os.environ["GEMINI_API_KEY"])
 
 SYSTEM_PROMPT = """You are a concise diagnostic assistant for HP Metal Jet S100 industrial 3D metal printers.
 You receive real-time sensor data and answer operator questions in 1-3 short sentences.
@@ -162,6 +162,10 @@ No data         → "No data available for [X]. Check that the printer ID is cor
 """
 
 
+_sql_model = genai.GenerativeModel(model_name="gemini-2.0-flash", system_instruction=_SQL_SYSTEM)
+_answer_model = genai.GenerativeModel(model_name="gemini-2.0-flash", system_instruction=_ANSWER_SYSTEM)
+
+
 def _extract_sql(text: str) -> str:
     text = text.strip()
     # strip markdown code fences if the model adds them anyway
@@ -180,20 +184,11 @@ def chat():
         return jsonify({"error": "prompt is required"}), 400
 
     # ── Step 1: ask LLM to generate a SQL query for the needed data ──────────
-    sql_resp = claude.messages.create(
-        model="claude-haiku-4-5-20251001",
-        max_tokens=400,
-        system=_SQL_SYSTEM,
-        messages=[{
-            "role": "user",
-            "content": (
-                f"Database schema:\n{_SQL_SYSTEM}\n\n"
-                f"printer_id: {printer_id}\n"
-                f"User question: {prompt}"
-            ),
-        }],
+    sql_resp = _sql_model.generate_content(
+        f"Database schema:\n{_SQL_SYSTEM}\n\nprinter_id: {printer_id}\nUser question: {prompt}",
+        generation_config=genai.GenerationConfig(max_output_tokens=400),
     )
-    sql = _extract_sql(sql_resp.content[0].text)
+    sql = _extract_sql(sql_resp.text)
     print(f"\n[SQL] {sql}\n")
 
     # ── Step 2: run the query against Supabase ───────────────────────────────
@@ -210,23 +205,15 @@ def chat():
     print(f"[DB RESULT] {db_text[:500]}\n")
 
     # ── Step 3: ask LLM to answer using the retrieved data ───────────────────
-    answer_resp = claude.messages.create(
-        model="claude-haiku-4-5-20251001",
-        max_tokens=300,
-        system=_ANSWER_SYSTEM,
-        messages=[{
-            "role": "user",
-            "content": (
-                f"Data fetched from database:\n{db_text}\n\n"
-                f"Operator question: {prompt}"
-            ),
-        }],
+    answer_resp = _answer_model.generate_content(
+        f"Data fetched from database:\n{db_text}\n\nOperator question: {prompt}",
+        generation_config=genai.GenerationConfig(max_output_tokens=300),
     )
 
-    answer = answer_resp.content[0].text
-    print(f"[CLAUDE ANSWER] {answer}\n")
+    answer = answer_resp.text
+    print(f"[GEMINI ANSWER] {answer}\n")
     return jsonify({"text": answer})
 
 
 if __name__ == "__main__":
-    app.run(host="0.0.0.0", port=7860, debug=False)
+    app.run(host="0.0.0.0", port=7860, debug=False)