Update gradio_new_rag_app.py

gradio_new_rag_app.py

@@ -1,138 +1,147 @@
-#!/usr/bin/env python3
-"""
-Gradio UI for the structured RAG AI Search.
-
-HF Spaces usage:
-- HF runs app.py which imports make_app() from this module.
-- Do NOT call demo.launch() on import.
-"""
-
-import json
-import os
-import pandas as pd
-import gradio as gr
-
-from rag_treatment_app import RAGTreatmentSearchApp
-
-
-APP_TITLE = "Aesthetic AI Search (Structured RAG - Region → Sub-Zone → Issue → Type)"
-
-
-def format_answer_markdown(out: dict) -> str:
-    if not isinstance(out, dict):
-        return "No output."
-
-    answer_md = (out.get("answer_md") or "").strip()
-    sources = out.get("sources") or []
-
-    md = []
-    md.append(answer_md if answer_md else "No answer generated.")
-
-    if sources:
-        md.append("\n---\n")
-        md.append("## Sources")
-        for u in sources[:20]:
-            md.append(f"- {u}")
-
-    return "\n".join(md).strip()
-
-
-def build_debug_table(data: dict):
-    rows = []
-    dbg = (data or {}).get("_debug") or {}
-    for c in dbg.get("candidates") or []:
-        rows.append({
-            "procedure": c.get("procedure"),
-            "similarity": c.get("similarity"),
-            "type": c.get("type"),
-            "region": c.get("region"),
-            "sub_zone": c.get("sub_zone"),
-        })
-    return pd.DataFrame(rows)
-
-
-def make_app():
-    rag = RAGTreatmentSearchApp(
-        excel_path=os.getenv("DB_XLSX", "database.xlsx"),
-        embeddings_cache_path=os.getenv("EMB_CACHE", "treatment_embeddings.pkl"),
-    )
-
-    regions = rag.get_regions()
-
-    def subzones_for_region(region):
-        if not region:
-            return []
-        return rag.get_sub_zones(region)
-
-    def run_search(region, sub_zone, issue, pref, retrieval_k, final_k, show_debug):
-        out = rag.answer(
-            region=region,
-            sub_zone=sub_zone,
-            type_choice=pref,
-            issue_text=issue,
-            retrieval_k=int(retrieval_k),
-            final_k=int(final_k),
-        )
-        md = format_answer_markdown(out)
-        dbg_df = build_debug_table(out) if show_debug else pd.DataFrame()
-        raw = json.dumps(out, ensure_ascii=False, indent=2)
-        return md, dbg_df, raw
-
-    with gr.Blocks(title=APP_TITLE) as demo:
-        gr.Markdown(f"# {APP_TITLE}")
-
-        with gr.Row():
-            region = gr.Dropdown(
-                choices=regions,
-                label="Region (Body part)",
-                value=regions[0] if regions else None
-            )
-            sub_zone = gr.Dropdown(
-                choices=subzones_for_region(regions[0] if regions else ""),
-                label="Sub-Zone",
-                interactive=True
-            )
-
-        def _update_subzones(r):
-            return gr.Dropdown(choices=subzones_for_region(r), value=None)
-
-        region.change(_update_subzones, inputs=[region], outputs=[sub_zone])
-
-        issue = gr.Textbox(lines=3, label="Describe your issue/problem (free text, multilingual supported)")
-        pref = gr.Radio(
-            ["Surgical Treatment", "Non-surgical Treatment", "Both"],
-            value="Both",
-            label="Treatment preference"
-        )
-
-        with gr.Row():
-            retrieval_k = gr.Slider(5, 30, value=12, step=1, label="Retrieval candidates (semantic top-K)")
-            final_k = gr.Slider(3, 10, value=5, step=1, label="Final recommendations (LLM top-K)")
-
-        show_debug = gr.Checkbox(value=False, label="Show debug (semantic candidates)")
-        run_btn = gr.Button("Run AI Search")
-
-        md_out = gr.Markdown()
-        dbg_out = gr.Dataframe(label="Debug: Semantic candidates")
-        raw_json = gr.Code(label="Raw JSON")
-
-        run_btn.click(
-            run_search,
-            inputs=[region, sub_zone, issue, pref, retrieval_k, final_k, show_debug],
-            outputs=[md_out, dbg_out, raw_json],
-        )
-
-        gr.Markdown(
-            """
-### Deployment note (Hugging Face)
-This Space runs an embedded open-source Transformers model (CPU).
-Web evidence may be unavailable at times due to rate limits or blocked requests; the app will still return database-driven results.
-"""
-        )
-
-    return demo
-
-
-if __name__ == "__main__":
-    demo = make_app()
-    demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)
+#!/usr/bin/env python3
+"""
+Gradio UI for the structured RAG AI Search.
+
+Hugging Face Spaces usage:
+- app.py imports make_app() and exposes `demo` at top-level.
+- DO NOT call demo.launch() in this file.
+"""
+
+import json
+import os
+import pandas as pd
+import gradio as gr
+
+from rag_treatment_app import RAGTreatmentSearchApp
+
+
+APP_TITLE = "Aesthetic AI Search (Structured RAG - Region → Sub-Zone → Issue → Type)"
+
+
+def format_answer_markdown(out: dict) -> str:
+    if not isinstance(out, dict):
+        return "No output."
+
+    answer_md = (out.get("answer_md") or "").strip()
+    sources = out.get("sources") or []
+
+    md = []
+    md.append(answer_md if answer_md else "No answer generated.")
+
+    if sources:
+        md.append("\n---\n")
+        md.append("## Sources")
+        for u in sources[:20]:
+            md.append(f"- {u}")
+
+    return "\n".join(md).strip()
+
+
+def build_debug_table(data: dict):
+    rows = []
+    dbg = (data or {}).get("_debug") or {}
+    for c in dbg.get("candidates") or []:
+        rows.append(
+            {
+                "procedure": c.get("procedure"),
+                "similarity": c.get("similarity"),
+                "type": c.get("type"),
+                "region": c.get("region"),
+                "sub_zone": c.get("sub_zone"),
+            }
+        )
+    return pd.DataFrame(rows)
+
+
+def make_app():
+    # Load paths from env (HF-friendly)
+    db_path = os.getenv("DB_XLSX", "database.xlsx")
+    emb_path = os.getenv("EMB_CACHE", "treatment_embeddings.pkl")
+
+    rag = RAGTreatmentSearchApp(
+        excel_path=db_path,
+        embeddings_cache_path=emb_path,
+    )
+
+    regions = rag.get_regions()
+
+    def subzones_for_region(region):
+        if not region:
+            return []
+        return rag.get_sub_zones(region)
+
+    def run_search(region, sub_zone, issue, pref, retrieval_k, final_k, show_debug):
+        out = rag.answer(
+            region=region,
+            sub_zone=sub_zone,
+            type_choice=pref,
+            issue_text=issue,
+            retrieval_k=int(retrieval_k),
+            final_k=int(final_k),
+        )
+        md = format_answer_markdown(out)
+        dbg_df = build_debug_table(out) if show_debug else pd.DataFrame()
+        raw = json.dumps(out, ensure_ascii=False, indent=2)
+        return md, dbg_df, raw
+
+    with gr.Blocks(title=APP_TITLE) as demo:
+        gr.Markdown(f"# {APP_TITLE}")
+
+        with gr.Row():
+            region = gr.Dropdown(
+                choices=regions,
+                label="Region (Body part)",
+                value=regions[0] if regions else None,
+            )
+            sub_zone = gr.Dropdown(
+                choices=subzones_for_region(regions[0] if regions else ""),
+                label="Sub-Zone",
+                interactive=True,
+            )
+
+        def _update_subzones(r):
+            return gr.Dropdown(choices=subzones_for_region(r), value=None)
+
+        region.change(_update_subzones, inputs=[region], outputs=[sub_zone])
+
+        issue = gr.Textbox(
+            lines=3,
+            label="Describe your issue/problem (free text, multilingual supported)",
+        )
+
+        pref = gr.Radio(
+            ["Surgical Treatment", "Non-surgical Treatment", "Both"],
+            value="Both",
+            label="Treatment preference",
+        )
+
+        with gr.Row():
+            retrieval_k = gr.Slider(
+                5, 30, value=12, step=1, label="Retrieval candidates (semantic top-K)"
+            )
+            final_k = gr.Slider(
+                3, 10, value=5, step=1, label="Final recommendations (LLM top-K)"
+            )
+
+        show_debug = gr.Checkbox(value=False, label="Show debug (semantic candidates)")
+        run_btn = gr.Button("Run AI Search")
+
+        md_out = gr.Markdown()
+        dbg_out = gr.Dataframe(label="Debug: Semantic candidates")
+        raw_json = gr.Code(label="Raw JSON")
+
+        run_btn.click(
+            run_search,
+            inputs=[region, sub_zone, issue, pref, retrieval_k, final_k, show_debug],
+            outputs=[md_out, dbg_out, raw_json],
+        )
+
+        gr.Markdown(
+            """
+### Deployment note (Hugging Face)
+This Space runs an embedded open-source Transformers model (CPU).
+Web evidence may be unavailable at times due to rate limits or blocked requests; the app will still return database-driven results.
+"""
+        )
+
+    return demo