Upload backend.py with huggingface_hub

backend.py

@@ -6,8 +6,8 @@ Flow:
 1. Receive Croatian discharge letter from doctor
 2. Translate to English (Gemini 3.1 Flash Lite)
 3. Run concurrent error-detection analysis:
-   - Qwen 3 32B (via Groq)
-   - Llama 4 Scout (via Groq)
+   - DeepSeek Reasoner (via DeepSeek API)
+   - GPT-OSS-120B (via Groq)
 4. Parse structured output and return errors + suggestions
 """
 
@@ -19,6 +19,7 @@ from dataclasses import dataclass, field
 from typing import Optional
 from dotenv import load_dotenv
 from google import genai
+from openai import OpenAI
 from groq import Groq
 
 load_dotenv(dotenv_path=os.path.join(os.path.dirname(__file__), ".env"))
@@ -32,6 +33,13 @@ def get_gemini_client() -> genai.Client:
     return genai.Client(api_key=key)
 
 
+def get_deepseek_client() -> OpenAI:
+    return OpenAI(
+        api_key=os.environ.get("DEEPSEEK_API_KEY"),
+        base_url="https://api.deepseek.com",
+    )
+
+
 def get_groq_client() -> Groq:
     return Groq(api_key=os.environ.get("GROQ_API_KEY_OSS"))
 
@@ -176,55 +184,53 @@ def parse_model_json(raw: str) -> dict:
 # Model calls
 # ---------------------------------------------------------------------------
 
-GROQ_MODEL_A = "qwen/qwen3-32b"
-GROQ_MODEL_B = "meta-llama/llama-4-scout-17b-16e-instruct"
+def _parse_to_result(model_label: str, raw: str, latency: float) -> ModelResult:
+    parsed = parse_model_json(raw)
+    errors = [
+        ParsedError(
+            description=e.get("description", ""),
+            category=e.get("category", "other"),
+            severity=e.get("severity", "medium"),
+            quote=e.get("quote", ""),
+        )
+        for e in parsed.get("errors", [])
+    ]
+    suggestions = [
+        ParsedSuggestion(
+            description=s.get("description", ""),
+            category=s.get("category", "other"),
+        )
+        for s in parsed.get("suggestions", [])
+    ]
+    return ModelResult(
+        model_name=model_label,
+        raw_response=raw,
+        errors=errors,
+        suggestions=suggestions,
+        summary=parsed.get("summary", ""),
+        success=True,
+        latency_seconds=round(latency, 2),
+    )
 
 
-def _call_groq_model(model_id: str, model_label: str, clinical_text: str) -> ModelResult:
+def call_model_a(clinical_text: str) -> ModelResult:
+    """DeepSeek Reasoner via DeepSeek API."""
     start = time.time()
     try:
-        client = get_groq_client()
+        client = get_deepseek_client()
         response = client.chat.completions.create(
-            model=model_id,
+            model="deepseek-reasoner",
             messages=[
                 {"role": "system", "content": ERROR_CHECK_SYSTEM_PROMPT},
                 {"role": "user", "content": ERROR_CHECK_USER_PROMPT.format(clinical_text=clinical_text)},
             ],
-            temperature=0.2,
             max_tokens=4096,
         )
         raw = response.choices[0].message.content
-        latency = time.time() - start
-
-        parsed = parse_model_json(raw)
-        errors = [
-            ParsedError(
-                description=e.get("description", ""),
-                category=e.get("category", "other"),
-                severity=e.get("severity", "medium"),
-                quote=e.get("quote", ""),
-            )
-            for e in parsed.get("errors", [])
-        ]
-        suggestions = [
-            ParsedSuggestion(
-                description=s.get("description", ""),
-                category=s.get("category", "other"),
-            )
-            for s in parsed.get("suggestions", [])
-        ]
-        return ModelResult(
-            model_name=model_label,
-            raw_response=raw,
-            errors=errors,
-            suggestions=suggestions,
-            summary=parsed.get("summary", ""),
-            success=True,
-            latency_seconds=round(latency, 2),
-        )
+        return _parse_to_result("DeepSeek Reasoner", raw, time.time() - start)
     except Exception as exc:
         return ModelResult(
-            model_name=model_label,
+            model_name="DeepSeek Reasoner",
             raw_response="",
             success=False,
             error_message=str(exc),
@@ -232,12 +238,30 @@ def _call_groq_model(model_id: str, model_label: str, clinical_text: str) -> Mod
         )
 
 
-def call_model_a(clinical_text: str) -> ModelResult:
-    return _call_groq_model(GROQ_MODEL_A, "Qwen 3 32B", clinical_text)
-
-
 def call_model_b(clinical_text: str) -> ModelResult:
-    return _call_groq_model(GROQ_MODEL_B, "Llama 4 Scout", clinical_text)
+    """GPT-OSS-120B via Groq."""
+    start = time.time()
+    try:
+        client = get_groq_client()
+        response = client.chat.completions.create(
+            model="openai/gpt-oss-120b",
+            messages=[
+                {"role": "system", "content": ERROR_CHECK_SYSTEM_PROMPT},
+                {"role": "user", "content": ERROR_CHECK_USER_PROMPT.format(clinical_text=clinical_text)},
+            ],
+            temperature=0.2,
+            max_tokens=4096,
+        )
+        raw = response.choices[0].message.content
+        return _parse_to_result("GPT-OSS-120B", raw, time.time() - start)
+    except Exception as exc:
+        return ModelResult(
+            model_name="GPT-OSS-120B",
+            raw_response="",
+            success=False,
+            error_message=str(exc),
+            latency_seconds=round(time.time() - start, 2),
+        )
 
 
 # ---------------------------------------------------------------------------
@@ -270,6 +294,9 @@ def run_error_check(croatian_text: str) -> AnalysisResponse:
 # ---------------------------------------------------------------------------
 
 if __name__ == "__main__":
+    import sys, io
+    sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding="utf-8", errors="replace")
+
     sample = """Bolesnik 68 godina, dolazi zbog bolova u prsištu.
     Dijagnoza: STEMI prednje stijenke.
     Terapija: Aspirin 100mg, Klopidogrel 75mg, Ramipril 5mg, Atorvastatin 40mg.