Update app.py

app.py

@@ -11,20 +11,45 @@ TOP_K_DEF = int(os.getenv("TOP_K", "3"))
 X_TITLE   = os.getenv("X_TITLE", "Undersea Species API")
 REFERER   = os.getenv("REFERER", "")
 
-# ===== System prompt (EN only; strict JSON) =====
+# ===== System Prompt =====
 SYSTEM_PROMPT = (
     "You are a marine-biology assistant. Return ONLY a valid JSON object with this schema:\n"
-    '{"candidates":[{"common_name": string, "scientific_name": string, "confidence": number, "blurb": string}]}\n'
+    '{"candidates":[{'
+    '"common_name": string, '
+    '"scientific_name": string, '
+    '"confidence": number, '
+    '"blurb": string, '
+    '"fun_facts": [string], '
+    '"summary": string, '
+    '"why_it_matters": string, '
+    '"conservation_status": string, '
+    '"status_source": string, '
+    '"local_guidelines": [string], '
+    '"species_type": string, '
+    '"occurrence_data": string, '
+    '"taxonomy": string, '
+    '"depth_range": string, '
+    '"observation_period": string'
+    '}]} \n'
     "Rules:\n"
-    "- Up to {TOP_K} items.\n"
+    f"- Up to {{TOP_K}} items.\n"
     "- confidence is a float in [0,1].\n"
-    "- If species is uncertain, use a higher-rank taxon (order/class) and mention uncertainty in blurb.\n"
-    "- Output JSON only. No explanations, no markdown, no code fences, no nested JSON inside strings."
+    "- fun_facts: 1–3 short, engaging facts.\n"
+    "- summary: 2–3 sentences, must include habitat, diet, role.\n"
+    "- why_it_matters: 1 short highlight (e.g., 'Indicator of reef health').\n"
+    "- conservation_status: use IUCN categories (e.g., 'Endangered', 'Vulnerable').\n"
+    "- status_source: always 'IUCN Red List' if from IUCN.\n"
+    "- local_guidelines: safety/protection rules for divers.\n"
+    "- species_type: fish, mammal, reptile, coral, etc.\n"
+    "- occurrence_data: how often it’s sighted.\n"
+    "- taxonomy: scientific classification.\n"
+    "- depth_range: typical depth (e.g., '5–30m').\n"
+    "- observation_period: seasonal or yearly occurrence.\n"
+    "- Output JSON only. No markdown, no explanations."
 )
 
 # ===== Helpers =====
 def pil_from_any(img_input: Any) -> Image.Image:
-    """Accepts Gradio PIL, URL, data-url, or base64 string."""
     if isinstance(img_input, Image.Image):
         return img_input.convert("RGB")
     if isinstance(img_input, str):
@@ -36,7 +61,6 @@ def pil_from_any(img_input: Any) -> Image.Image:
             r = requests.get(img_input, timeout=30)
             r.raise_for_status()
             return Image.open(io.BytesIO(r.content)).convert("RGB")
-        # bare base64
         try:
             data = base64.b64decode(img_input)
             return Image.open(io.BytesIO(data)).convert("RGB")
@@ -51,20 +75,35 @@ def pil_to_data_url(img: Image.Image) -> str:
     return f"data:image/png;base64,{b64}"
 
 def coerce_json(text: str) -> Dict:
-    """
-    Best-effort: parse JSON, or extract the last {...} block.
-    Always return dict with 'candidates' array.
-    """
+    default = {
+        "candidates": [
+            {
+                "common_name": "unknown",
+                "scientific_name": "unknown",
+                "confidence": 0.0,
+                "blurb": "empty response",
+                "fun_facts": [],
+                "summary": "No data",
+                "why_it_matters": "No data",
+                "conservation_status": "Unknown",
+                "status_source": "Unknown",
+                "local_guidelines": [],
+                "species_type": "unknown",
+                "occurrence_data": "No data",
+                "taxonomy": "No data",
+                "depth_range": "Unknown",
+                "observation_period": "Unknown"
+            }
+        ]
+    }
     if not text:
-        return {"candidates":[{"common_name":"unknown","scientific_name":"unknown","confidence":0.0,"blurb":"empty response"}]}
-    # try direct
+        return default
     try:
         obj = json.loads(text)
         if isinstance(obj, dict) and "candidates" in obj:
             return obj
     except Exception:
         pass
-    # try to extract last {...}
     import re
     m = re.findall(r"\{(?:[^{}]|(?R))*\}", text, flags=re.DOTALL)
     if m:
@@ -75,31 +114,27 @@ def coerce_json(text: str) -> Dict:
                     return obj
             except Exception:
                 continue
-    # fallback
-    return {"candidates":[{"common_name":"unknown","scientific_name":"unknown","confidence":0.0,"blurb":text[:400]}]}
+    default["candidates"][0]["blurb"] = text[:400]
+    return default
 
-# ===== OpenRouter call (OpenAI-compatible chat.completions) =====
+# ===== API Call =====
 def call_openrouter(img: Image.Image, top_k: int) -> Dict:
     if not API_KEY:
-        return {"candidates":[{"common_name":"unknown","scientific_name":"unknown","confidence":0.0,"blurb":"API_KEY not set"}]}
-
+        return coerce_json("API_KEY not set")
     headers = {
         "Authorization": f"Bearer {API_KEY}",
         "Content-Type": "application/json",
     }
-    if X_TITLE:
-        headers["X-Title"] = X_TITLE
-    if REFERER:
-        headers["HTTP-Referer"] = REFERER
+    if X_TITLE: headers["X-Title"] = X_TITLE
+    if REFERER: headers["HTTP-Referer"] = REFERER
 
     data_url = pil_to_data_url(img)
     sys_prompt = SYSTEM_PROMPT.replace("{TOP_K}", str(top_k))
-
     payload = {
         "model": MODEL_ID,
         "response_format": {"type": "json_object"},
-        "temperature": 0,            # strict output
-        "max_tokens": 700,
+        "temperature": 0,
+        "max_tokens": 900,
         "messages": [
             {"role": "system", "content": sys_prompt},
             {"role": "user", "content": [
@@ -110,7 +145,7 @@ def call_openrouter(img: Image.Image, top_k: int) -> Dict:
     }
 
     last_err = ""
-    for i in range(2):  # light retry
+    for i in range(2):
         try:
             resp = requests.post(API_URL, headers=headers, json=payload, timeout=90)
             if resp.ok:
@@ -120,11 +155,10 @@ def call_openrouter(img: Image.Image, top_k: int) -> Dict:
             last_err = f"{resp.status_code}: {resp.text[:600]}"
         except Exception as e:
             last_err = str(e)[:600]
-        time.sleep(0.4 * (i + 1))
-
-    return {"candidates":[{"common_name":"unknown","scientific_name":"unknown","confidence":0.0,"blurb":f"Upstream error: {last_err}"}]}
+        time.sleep(0.5 * (i + 1))
+    return coerce_json(f"Upstream error: {last_err}")
 
-# ===== Gradio UI & API =====
+# ===== Gradio UI =====
 def classify(image, top_k=TOP_K_DEF):
     pil = pil_from_any(image)
     res = call_openrouter(pil, int(top_k))
@@ -135,10 +169,10 @@ def classify(image, top_k=TOP_K_DEF):
         "timestamp": time.time()
     }
 
-with gr.Blocks(title="Underwater Species (OpenRouter + GPT-4o-mini)") as demo:
-    gr.Markdown("## Underwater Species Recognizer (OpenRouter → Vision LLM)\n"
+with gr.Blocks(title="Underwater Species Recognition (Educational + Conservation)") as demo:
+    gr.Markdown("## 🌊 Underwater Species Recognizer (AI + Conservation)\n"
                 f"**Model:** `{MODEL_ID}`  \n"
-                "Set `API_KEY`, `API_URL`, `MODEL_ID` in Space Secrets.")
+                "Upload an underwater photo to identify species, learn facts, and see conservation status.")
     with gr.Row():
         with gr.Column(scale=1):
             img = gr.Image(type="pil", label="Underwater image")