Upload app.py

app.py

@@ -1,11 +1,12 @@
-# app.py — DiagStudio AI (Streamlit frontend for the FastAPI backend with a single composite image)
-# UI is end-user friendly (English). It calls your private backend's /analyze and /analyze_zip.
-# Notes:
-# - Set CARDIAG_AI_API (or AI_LIGHTBOX_API / LUXFIT_API) to your private FastAPI base URL.
-# - Optional: set CARDIAG_AI_TOKEN (or AI_LIGHTBOX_TOKEN / LUXFIT_TOKEN) for auth header passthrough.
-# - We disable Streamlit's telemetry writes and force a writable HOME to avoid /.streamlit permission issues.
+# app.py — DiagStudio AI (Streamlit frontend)
+# Works with the FastAPI backend you just built (single composite image + JSON mode).
+# Configure the backend URL via:
+#   CARDIAG_AI_API  (or AI_LIGHTBOX_API / LUXFIT_API)
+# Optional auth header passthrough:
+#   CARDIAG_AI_TOKEN (or AI_LIGHTBOX_TOKEN / LUXFIT_TOKEN)
 
 import os, io, json, time, base64, requests, hashlib
+# Prevent /.streamlit permission issues on HF Spaces/containers
 os.environ.setdefault("HOME", "/tmp")
 os.environ.setdefault("STREAMLIT_BROWSER_GATHER_USAGE_STATS", "false")
 os.environ.setdefault("STREAMLIT_SERVER_HEADLESS", "true")
@@ -34,7 +35,7 @@ HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"} if HF_TOKEN else {}
 
 st.set_page_config(page_title="DiagStudio AI", layout="wide")
 st.title("DiagStudio AI")
-st.caption("Upload a short engine recording and get a clear diagnosis with visuals and notes.")
+st.caption("Upload a short engine recording and get a clear diagnosis with visuals and next steps.")
 
 # ================= Helpers =================
 @st.cache_data(show_spinner=False, ttl=20)
@@ -65,7 +66,7 @@ def _post_analyze(audio_file, rpm_file, sr_target, n_fft, hop, env_band, run_llm
         "env_band": env_band,
         "run_llm": "true" if run_llm else "false",
         "extra_text": notes or "",
-        # cache-bypass hint for proxies/CDNs (backend already sets no-store)
+        # cache-bypass hint for intermediaries (backend already sets no-store)
         "nonce": str(time.time())
     }
     r = requests.post(f"{API_BASE}/analyze", headers=HEADERS, files=files, data=data, timeout=240)
@@ -98,7 +99,7 @@ with st.sidebar:
     rpm_up = st.file_uploader("Optional RPM CSV (time_sec,rpm)", type=["csv"])
 
     st.header("Options")
-    run_llm = st.toggle("Explain with AI", value=True, help="Get a readable diagnosis and next steps.")
+    run_llm = st.toggle("Explain with AI", value=True, help="Adds an easy-to-read diagnosis and next steps.")
     user_notes = st.text_area("Context (optional)", placeholder="Vehicle, engine type, driving condition, symptoms…")
 
     with st.expander("Advanced (defaults are fine)"):
@@ -114,7 +115,7 @@ with st.sidebar:
         st.warning("Could not reach the backend. Verify your private Space URL.")
     analyze_btn = st.button("Analyze now", type="primary", disabled=(audio_up is None or not ok))
 
-# Clear old result if user picks a different file (prevents “same result” confusion)
+# ================= Clear old result on file change =================
 if "last_audio_id" not in st.session_state:
     st.session_state["last_audio_id"] = None
 if audio_up is not None:
@@ -144,26 +145,33 @@ res = st.session_state.get("last_result")
 if res:
     st.subheader("Result")
 
-    llm = res.get("llm_report") or {}
-    model_name = llm.get("model", "unknown")
-    usage = llm.get("usage") or {}
     params = res.get("params", {})
+    llm = res.get("llm_report") or {}
+
+    # Model name is known from backend params (JSON mode removes auto-added model field)
+    model_name = params.get("vision_model", "gpt-4o")
     st.caption(f"AI model: **{model_name}**")
-    if usage:
-        st.caption(f"Tokens — prompt: {usage.get('prompt_tokens')}, completion: {usage.get('completion_tokens')}")
+
+    # Show audio hash and basic features to prove input is unique
     if params.get("audio_id"):
         st.caption(f"Audio ID: **{params['audio_id']}**")
     if params.get("features"):
         feat = params["features"]
-        st.caption(
-            f"Features — rms: {feat.get('rms'):.4f}, zcr: {feat.get('zcr'):.4f}, "
-            f"centroid: {feat.get('centroid'):.1f} Hz, rolloff: {feat.get('rolloff'):.1f} Hz, "
-            f"flux: {feat.get('flux'):.2f}"
-        )
-    meta = llm.get("meta", {})
-    if meta:
+        try:
+            st.caption(
+                f"Features — rms: {feat.get('rms'):.4f}, zcr: {feat.get('zcr'):.4f}, "
+                f"centroid: {feat.get('centroid'):.1f} Hz, rolloff: {feat.get('rolloff'):.1f} Hz, "
+                f"flux: {feat.get('flux'):.2f}"
+            )
+        except Exception:
+            st.caption(f"Features — {feat}")
+
+    # If model returned meta, show it (proof it saw the image)
+    if isinstance(llm, dict) and llm.get("meta"):
+        meta = llm["meta"]
         st.caption(f"Vision meta — seen_images: {meta.get('seen_images')}, views: {meta.get('views')}")
 
+    # Layout
     col1, col2 = st.columns([1, 1])
     diag_list = llm.get("diagnosis", [])
     notes = llm.get("notes", "")
@@ -171,10 +179,12 @@ if res:
 
     with col1:
         st.markdown("**Likely issues**")
-        if diag_list:
+        if isinstance(diag_list, list) and len(diag_list) > 0:
             for item in diag_list:
-                st.text(str(item.get("label", "Issue")))
-                st.progress(_nice_conf(item.get("likelihood", 0)))
+                label = str(item.get("label", "Issue"))
+                conf = _nice_conf(item.get("likelihood", 0))
+                st.text(label)
+                st.progress(conf)
                 if item.get("evidence"):
                     st.caption(item["evidence"])
         else:
@@ -189,8 +199,15 @@ if res:
             st.write(next_tests)
 
     st.subheader("Visual checks")
-    tabs = st.tabs(["Composite", "Energy over time", "Perceptual view", "Impact pattern", "Signal envelope", "Order view"])
-    imgs = res.get("images", {})
+    tabs = st.tabs([
+        "Composite",
+        "Energy over time",
+        "Perceptual view",
+        "Impact pattern",
+        "Signal envelope",
+        "Order view",
+    ])
+    imgs = res.get("images", {}) if isinstance(res.get("images", {}), dict) else {}
 
     with tabs[0]:
         if imgs.get("composite"):
@@ -224,7 +241,12 @@ if res:
     if st.button("Prepare ZIP"):
         try:
             zbytes = _post_analyze_zip(audio_up, rpm_up, sr_target, n_fft, hop, env_band, run_llm, user_notes)
-            st.download_button("Download results (ZIP)", data=zbytes, file_name="diagstudio_results.zip", mime="application/zip")
+            st.download_button(
+                "Download results (ZIP)",
+                data=zbytes,
+                file_name="diagstudio_results.zip",
+                mime="application/zip"
+            )
         except Exception as e:
             st.error(f"ZIP build failed: {e}")
 else: