Update app.py

app.py

@@ -67,32 +67,54 @@ def fetch_mp3_by_unique_id(unique_id: str) -> Tuple[str, str]:
 # ─────────────────────────────────────────────────────────────────────────────
 PROMPT_TEMPLATES = {
     "simple": (
-        "You are a call-center conversation analyst for a medical clinic. From the call recording, provide a brief summary:\n"
-        "- Purpose of the call (appointment / results / complaint / billing / other).\n"
-        "- Patient intent and expectations.\n"
-        "- Outcome (booked / call-back / routed / unresolved).\n"
-        "- Next steps (owner and when).\n"
-        "- Patient emotion (1–5) and agent tone (1–5).\n"
-        "- Alerts: urgency/risks/privacy.\n\n"
+        "You are a call-center conversation analyst for a medical clinic. From the call recording, provide a brief summary:
+"
+        "- Purpose of the call (appointment / results / complaint / billing / other).
+"
+        "- Patient intent and expectations.
+"
+        "- Outcome (booked / call-back / routed / unresolved).
+"
+        "- Next steps (owner and when).
+"
+        "- Patient emotion (1–5) and agent tone (1–5).
+"
+        "- Alerts: urgency/risks/privacy.
+
+"
         "Keep it short (6–8 lines). End with a line: ‘Service quality rating: X/5’ and one sentence explaining the rating."
     ),
     "medium": (
-        "Act as a senior service analyst. Analyze the call using this structure:\n"
-        "1) Quick overview: reason for the call, intent, key facts, urgency (low/medium/high).\n"
-        "2) Call flow (2–4 bullets): what was asked/answered, where friction occurred.\n"
-        "3) Outcomes & tasks: concrete next actions for clinic/patient with timeframes.\n"
-        "4) Emotions & empathy: patient mood; agent empathy (0–5).\n"
-        "5) Procedural compliance: identity verification, disclosure of recording (if stated), no off-protocol medical advice, data accuracy.\n"
+        "Act as a senior service analyst. Analyze the call using this structure:
+"
+        "1) Quick overview: reason for the call, intent, key facts, urgency (low/medium/high).
+"
+        "2) Call flow (2–4 bullets): what was asked/answered, where friction occurred.
+"
+        "3) Outcomes & tasks: concrete next actions for clinic/patient with timeframes.
+"
+        "4) Emotions & empathy: patient mood; agent empathy (0–5).
+"
+        "5) Procedural compliance: identity verification, disclosure of recording (if stated), no off-protocol medical advice, data accuracy.
+"
         "6) Quality rating (0–100) using rubric: greeting, verification, accuracy, empathy, issue resolution (each 0–20)."
     ),
     "detailed": (
-        "You are a quality & operations analyst. Provide an in-depth analysis:\n"
-        "A) Segmentation: split the call into stages with approximate timestamps (if available) and roles (Patient/Agent).\n"
-        "B) Structured data for booking: full name (if stated), date of birth, phone, symptoms/complaints (list), onset/duration, possible pain level 0–10 (if mentioned), required specialist/service, preferred time windows, constraints.\n"
-        "C) Triage & risks: class (routine/urgent/emergency), red flags, whether immediate escalation is needed.\n"
-        "D) Compliance audit: identity/privacy checks, recording disclosure, consent to data processing, booking policies.\n"
-        "E) Conversation metrics: talk ratio (agent/patient), interruptions, long pauses, notable keywords.\n"
-        "F) Coaching for the agent: 3–5 concrete improvements with sample phrasing.\n\n"
+        "You are a quality & operations analyst. Provide an in-depth analysis:
+"
+        "A) Segmentation: split the call into stages with approximate timestamps (if available) and roles (Patient/Agent).
+"
+        "B) Structured data for booking: full name (if stated), date of birth, phone, symptoms/complaints (list), onset/duration, possible pain level 0–10 (if mentioned), required specialist/service, preferred time windows, constraints.
+"
+        "C) Triage & risks: class (routine/urgent/emergency), red flags, whether immediate escalation is needed.
+"
+        "D) Compliance audit: identity/privacy checks, recording disclosure, consent to data processing, booking policies.
+"
+        "E) Conversation metrics: talk ratio (agent/patient), interruptions, long pauses, notable keywords.
+"
+        "F) Coaching for the agent: 3–5 concrete improvements with sample phrasing.
+
+"
         "Deliver: (1) A short patient-chart summary (2–3 sentences). (2) A task table with columns: priority, owner, due."
     ),
 }
@@ -239,12 +261,26 @@ def _filter_calls(df: pd.DataFrame, t_from: Optional[str], t_to: Optional[str],
         if col:
             # exact match (case-insensitive)
             mask = mask & (work[col].astype(str).str.lower() == ct.lower())
-        else:
-            # if there is no calltype column, keep only time filter
-            pass
 
     return work.loc[mask].drop(columns=[c for c in ["_Start_dt", "_Start_time_sec"] if c in work.columns])
 
+
+def _validate_times(t_from: str, t_to: str) -> Tuple[bool, str, bool, bool]:
+    """Return (ok, message_html, from_valid, to_valid)."""
+    tf = _parse_hhmm(t_from)
+    tt = _parse_hhmm(t_to)
+    if tf is None and tt is None:
+        return False, "<span style='color:#c00'>❌ Памылка: няправільны фармат часу ў абодвух палях (выкарыстайце HH:MM).</span>", False, False
+    if tf is None:
+        return False, "<span style='color:#c00'>❌ Памылка: поле <b>Time from</b> павінна быць у фармаце HH:MM.</span>", False, True
+    if tt is None:
+        return False, "<span style='color:#c00'>❌ Памылка: поле <b>Time to</b> павінна быць у фармаце HH:MM.</span>", True, False
+    return True, "", True, True
+
+
+def _recording_url(unique_id: str) -> str:
+    return f"{BASE_URL}/calllogs/{CLIENT_ID}/{unique_id}"
+
 # ─────────────────────────────────────────────────────────────────────────────
 # Gradio handlers
 # ─────────────────────────────────────────────────────────────────────────────
@@ -264,7 +300,8 @@ def ui_fetch_calls(date_str: str):
             body = e.response.text[:800]
         except Exception:
             pass
-        return pd.DataFrame(), gr.update(choices=[], value=None), f"HTTP error: {e}\n{body}"
+        return pd.DataFrame(), gr.update(choices=[], value=None), f"HTTP error: {e}
+{body}"
     except Exception as e:
         return pd.DataFrame(), gr.update(choices=[], value=None), f"Load error: {e}"
 
@@ -279,7 +316,7 @@ def ui_play_audio(selected_idx: Optional[int], df: pd.DataFrame):
     unique_id = str(row.get("UniqueId"))
     try:
         fpath = f"/tmp/call_{unique_id}.mp3"
-        url_used = f"{BASE_URL}/calllogs/{CLIENT_ID}/{unique_id}"
+        url_used = _recording_url(unique_id)
         # Download only if not exists (avoid re-fetch)
         if not os.path.exists(fpath) or os.path.getsize(fpath) == 0:
             fpath, url_used = fetch_mp3_by_unique_id(unique_id)
@@ -348,18 +385,28 @@ def ui_analyze(selected_idx: Optional[int], df: pd.DataFrame,
 
     # Call model
     try:
-        merged = f"""[SYSTEM INSTRUCTION: {sys_inst}]\n\n{prompt}"""
+        merged = f"""[SYSTEM INSTRUCTION: {sys_inst}]
+
+{prompt}"""
         resp = client.models.generate_content(model=model_name, contents=[uploaded_file, merged])
         text = getattr(resp, "text", None)
         if not text:
             return "Analysis finished but returned no text. Check model settings and file format."
-        return f"### Analysis result\n\n{text}"
+        # Add a listening link at the end
+        url = _recording_url(unique_id)
+        return f"### Analysis result
+
+{text}
+
+**Recording:** {url}"
     except Exception as e:
         # Try to attach more error details
         msg = str(e)
         try:
             if hasattr(e, "args") and e.args:
-                msg = msg + "\n\n" + str(e.args[0])
+                msg = msg + "
+
+" + str(e.args[0])
         except Exception:
             pass
         return f"Error during model call: {msg}"
@@ -372,11 +419,68 @@ def ui_analyze(selected_idx: Optional[int], df: pd.DataFrame,
             pass
 
 # ─────────────────────────────────────────────────────────────────────────────
-# NEW: Batch analysis with time interval + calltype filter
+# NEW: Batch helpers for dynamic calltype choices and preview
+# ─────────────────────────────────────────────────────────────────────────────
+
+def ui_batch_load(date_str: str, authed: bool):
+    """Fetch calls for the date, return DF, calltype choices, and status."""
+    if not authed:
+        return gr.update(), gr.update(), "🔒 Увядзіце пароль, каб атрымаць званкі.", gr.update(visible=True)
+
+    date_str = (date_str or "").strip() or _today_str()
+    try:
+        calls = fetch_calllogs(date_str)
+        df = pd.DataFrame(calls)
+    except Exception as e:
+        return gr.update(), gr.update(choices=[""], value=""), f"Не ўдалося загрузіць спіс: {e}", gr.update(visible=False)
+
+    if df.empty:
+        return gr.update(value=pd.DataFrame()), gr.update(choices=[""], value=""), "За гэты дзень званкоў не знойдзена.", gr.update(visible=False)
+
+    # Build calltype list automatically
+    col = _infer_calltype_col(df)
+    if col:
+        types = sorted(set(str(x).strip() for x in df[col].dropna().unique() if str(x).strip()))
+        choices = [""] + types  # allow empty = no filter
+    else:
+        choices = [""]
+
+    return (
+        df,  # store DF in state via gr.State
+        gr.update(choices=choices, value=""),
+        f"Загружана званкоў: {len(df)}. Абярыце фільтр і націсніце ‘Папярэдні прагляд’.",
+        gr.update(visible=False),
+    )
+
+
+def ui_batch_preview(df: pd.DataFrame, t_from: str, t_to: str, calltype: str):
+    ok, msg, from_ok, to_ok = _validate_times(t_from, t_to)
+    if not ok:
+        return pd.DataFrame(), f"{msg}", gr.update(info=("✓" if from_ok else "Format: HH:MM")), gr.update(info=("✓" if to_ok else "Format: HH:MM"))
+
+    if df is None or len(df) == 0:
+        return pd.DataFrame(), "Спачатку загрузіце спіс на дату.", gr.update(), gr.update()
+
+    # Filter
+    df_f = _filter_calls(df, t_from, t_to, calltype)
+
+    # Add recording URL column for quick listening
+    if not df_f.empty and "UniqueId" in df_f.columns:
+        df_f = df_f.copy()
+        df_f["RecordingURL"] = df_f["UniqueId"].astype(str).map(_recording_url)
+
+    count = len(df_f)
+    info = f"Фільтр: {t_from}–{t_to}{' | calltype=' + calltype if calltype else ''}. Знойдзена: <b>{count}</b>."
+    return df_f, info, gr.update(info=""), gr.update(info="")
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Batch analysis with time interval + calltype filter + URL in output
 # ─────────────────────────────────────────────────────────────────────────────
 
 def ui_analyze_calls_by_date(
     authed: bool,
+    df_loaded: pd.DataFrame,
     date_str: str,
     t_from: str,
     t_to: str,
@@ -387,9 +491,8 @@ def ui_analyze_calls_by_date(
     model_pref: str,
 ):
     """
-    Fetch calls for the given date, filter them by time window and calltype,
-    and analyze one-by-one with the same settings as the single-call tab.
-    Save a .txt file and show results on screen. Also show how many calls matched the filter.
+    Analyze filtered calls one-by-one. Uses the already loaded DF to allow
+    dynamic calltype dropdown and preview step. Adds listening URL to each block.
     """
     # Password gate
     if not authed:
@@ -400,30 +503,31 @@ def ui_analyze_calls_by_date(
             gr.update(visible=True),                       # show pwd group
         )
 
-    date_str = (date_str or "").strip() or _today_str()
-    t_from = (t_from or _DEF_T_FROM).strip()
-    t_to   = (t_to or _DEF_T_TO).strip()
-    calltype = (calltype or "").strip()
+    # Validate time inputs
+    ok, msg, from_ok, to_ok = _validate_times(t_from, t_to)
+    if not ok:
+        return (msg, None, "", gr.update(visible=False))
 
-    # 1) Fetch list for the date
-    try:
-        calls = fetch_calllogs(date_str)
-        df = pd.DataFrame(calls)
-    except Exception as e:
-        return (f"Не ўдалося загрузіць спіс: {e}", None, "", gr.update())
+    # Ensure DF is present; if not, fallback to fetch (safety)
+    if df_loaded is None or len(df_loaded) == 0:
+        try:
+            calls = fetch_calllogs((date_str or _today_str()).strip())
+            df_loaded = pd.DataFrame(calls)
+        except Exception as e:
+            return (f"Не ўдалося загрузіць спіс: {e}", None, "", gr.update())
 
-    if df.empty:
+    if df_loaded.empty:
         return ("За гэты дзень званкоў не знойдзена.", None, "", gr.update())
 
-    # 2) Filter by time + calltype
-    df_f = _filter_calls(df, t_from, t_to, calltype)
+    # Filter by time + calltype
+    df_f = _filter_calls(df_loaded, t_from, t_to, calltype)
     total = len(df_f)
 
     if total == 0:
         info = f"Па фільтру ({t_from}–{t_to}{' | ' + calltype if calltype else ''}) — 0 званкоў."
         return (info, None, info, gr.update())
 
-    # 3) Prepare model
+    # Prepare model
     if not _HAS_GENAI:
         return ("❌ google-genai library not found.", None, "", gr.update())
 
@@ -440,9 +544,16 @@ def ui_analyze_calls_by_date(
     sys_inst = _system_instruction(lang_code)
     prompt = _prepare_prompt(template_key, custom_prompt)
 
-    # 4) Iterate filtered calls
+    # Iterate filtered calls
     results_blocks: List[str] = [
-        f"## Вынікі\nДата: {date_str}\n\nФільтр: {t_from}–{t_to}{' | calltype=' + calltype if calltype else ''}\n\nЗнойдзена званкоў: {total}\n\n---"
+        f"## Вынікі
+Дата: {(date_str or _today_str()).strip()}
+
+Фільтр: {t_from}–{t_to}{' | calltype=' + calltype if calltype else ''}
+
+Знойдзена званкоў: {total}
+
+---"
     ]
     analyzed = 0
 
@@ -457,14 +568,25 @@ def ui_analyze_calls_by_date(
             if not os.path.exists(mp3_path) or os.path.getsize(mp3_path) == 0:
                 mp3_path, _ = fetch_mp3_by_unique_id(unique_id)
         except Exception as e:
+            url = _recording_url(unique_id)
             results_blocks.append(
-                f"### Call {i+1}\n"
-                f"- UniqueId: {unique_id}\n"
-                f"- Start: {row.get('Start','')}\n"
-                f"- CallerId: {row.get('CallerId','')}\n"
-                f"- Destination: {row.get('Destination','')}\n"
-                f"- Duration: {row.get('Duration','')}\n\n"
-                f"**Памылка загрузкі аўдыё:** {e}\n"
+                f"### Call {i+1}
+"
+                f"- UniqueId: {unique_id}
+"
+                f"- Start: {row.get('Start','')}
+"
+                f"- CallerId: {row.get('CallerId','')}
+"
+                f"- Destination: {row.get('Destination','')}
+"
+                f"- Duration: {row.get('Duration','')}
+"
+                f"- Recording: {url}
+
+"
+                f"**Памылка загрузкі аўдыё:** {e}
+"
                 "---"
             )
             continue
@@ -472,7 +594,9 @@ def ui_analyze_calls_by_date(
         # Upload + generate
         try:
             uploaded_file = client.files.upload(file=mp3_path)
-            merged = f"[SYSTEM INSTRUCTION: {sys_inst}]\n\n{prompt}"
+            merged = f"[SYSTEM INSTRUCTION: {sys_inst}]
+
+{prompt}"
             resp = client.models.generate_content(
                 model=model_name,
                 contents=[uploaded_file, merged],
@@ -487,15 +611,28 @@ def ui_analyze_calls_by_date(
             except Exception:
                 pass
 
-        # Result block
+        # Result block with listening URL
+        url = _recording_url(unique_id)
         block = (
-            f"### Call {i+1}\n"
-            f"- UniqueId: {unique_id}\n"
-            f"- Start: {row.get('Start','')}\n"
-            f"- CallerId: {row.get('CallerId','')}\n"
-            f"- Destination: {row.get('Destination','')}\n"
-            f"- Duration: {row.get('Duration','')}\n\n"
-            f"**Analysis:**\n\n{text}\n"
+            f"### Call {i+1}
+"
+            f"- UniqueId: {unique_id}
+"
+            f"- Start: {row.get('Start','')}
+"
+            f"- CallerId: {row.get('CallerId','')}
+"
+            f"- Destination: {row.get('Destination','')}
+"
+            f"- Duration: {row.get('Duration','')}
+"
+            f"- Recording: {url}
+
+"
+            f"**Analysis:**
+
+{text}
+"
             "---"
         )
         results_blocks.append(block)
@@ -505,9 +642,11 @@ def ui_analyze_calls_by_date(
         info = "Не атрымалася прааналізаваць ніводнага званка."
         return (info, None, info, gr.update())
 
-    # 5) Save to .txt and return
-    results_text = "\n\n".join(results_blocks)
-    fname = f"/tmp/batch_analysis_{date_str}_{t_from.replace(':','')}-{t_to.replace(':','')}_{int(time.time())}.txt"
+    # Save to .txt and return
+    results_text = "
+
+".join(results_blocks)
+    fname = f"/tmp/batch_analysis_{(date_str or _today_str()).strip()}_{t_from.replace(':','')}-{t_to.replace(':','')}_{int(time.time())}.txt"
     with open(fname, "w", encoding="utf-8") as f:
         f.write(results_text)
 
@@ -564,6 +703,9 @@ with gr.Blocks(title="Vochi CRM Call Logs (Gradio)") as demo:
     # Auth state (False by default)
     authed = gr.State(False)
 
+    # States for batch tab
+    batch_df_state = gr.State(pd.DataFrame())
+
     # Password "modal" (group shown on demand)
     with gr.Group(visible=False) as pwd_group:
         gr.Markdown("### 🔐 Увядзіце пароль")
@@ -593,14 +735,22 @@ with gr.Blocks(title="Vochi CRM Call Logs (Gradio)") as demo:
             analyze_btn = gr.Button("🧠 Analyze", variant="primary")
             analysis_md = gr.Markdown()
 
-        # NEW: пакетны аналіз з фільтрамі па часе і calltype
+        # Пакетны аналіз з фільтрамі па часе і calltype + прэв'ю
         with gr.Tab("Аналіз званкоў за дзень/час"):
             batch_date_inp = gr.Textbox(label="Date", value=_today_str(), scale=1)
 
             with gr.Row():
                 t_from_inp = gr.Textbox(label="Time from (HH:MM)", value=_DEF_T_FROM)
                 t_to_inp   = gr.Textbox(label="Time to (HH:MM)",   value=_DEF_T_TO)
-                calltype_inp = gr.Textbox(label="calltype (exact)", placeholder="e.g. inbound / outbound")
+                calltype_dd = gr.Dropdown(label="calltype", choices=[""], value="", info="Аўта-значэнні пасля загрузкі")
+
+            with gr.Row():
+                load_btn = gr.Button("Атрымаць спіс (для фільтраў)", variant="secondary")
+                preview_btn = gr.Button("Папярэдні прагляд", variant="secondary")
+                batch_btn = gr.Button("прааналізаваць", variant="primary")
+
+            preview_info_md = gr.Markdown()
+            preview_df = gr.Dataframe(value=pd.DataFrame(), label="Адфільтраваныя званкі", interactive=False)
 
             with gr.Row():
                 tpl_batch_dd = gr.Dropdown(choices=TPL_OPTIONS, value="simple", label="Template")
@@ -609,8 +759,6 @@ with gr.Blocks(title="Vochi CRM Call Logs (Gradio)") as demo:
 
             custom_batch_tb = gr.Textbox(label="Custom prompt", lines=8, visible=False)
 
-            batch_btn = gr.Button("прааналізаваць", variant="primary")
-
             batch_status_md = gr.Markdown()
             batch_results_md = gr.Markdown()
             batch_file_out = gr.File(label="Вынікі (.txt)")
@@ -630,7 +778,7 @@ with gr.Blocks(title="Vochi CRM Call Logs (Gradio)") as demo:
         outputs=[authed, status_fetch, pwd_group],
     )
 
-    # 3) Other interactions
+    # 3) Other interactions (single analysis)
     play_btn.click(ui_play_audio, inputs=[row_dd, calls_df], outputs=[url_html, audio_out, file_out, status_fetch])
     tpl_dd.change(ui_toggle_custom_prompt, inputs=[tpl_dd], outputs=[custom_prompt_tb])
     analyze_btn.click(
@@ -639,11 +787,27 @@ with gr.Blocks(title="Vochi CRM Call Logs (Gradio)") as demo:
         outputs=[analysis_md],
     )
 
-    # NEW: events for the new batch tab with filters
+    # 4) Batch tab interactions
     tpl_batch_dd.change(ui_toggle_custom_prompt, inputs=[tpl_batch_dd], outputs=[custom_batch_tb])
+
+    # Load list for date → populate state & calltype dropdown
+    load_btn.click(
+        ui_batch_load,
+        inputs=[batch_date_inp, authed],
+        outputs=[batch_df_state, calltype_dd, preview_info_md, pwd_group],
+    )
+
+    # Preview filtered table
+    preview_btn.click(
+        ui_batch_preview,
+        inputs=[batch_df_state, t_from_inp, t_to_inp, calltype_dd],
+        outputs=[preview_df, preview_info_md, t_from_inp, t_to_inp],
+    )
+
+    # Analyze filtered
     batch_btn.click(
         ui_analyze_calls_by_date,
-        inputs=[authed, batch_date_inp, t_from_inp, t_to_inp, calltype_inp, tpl_batch_dd, custom_batch_tb, lang_batch_dd, model_batch_dd],
+        inputs=[authed, batch_df_state, batch_date_inp, t_from_inp, t_to_inp, calltype_dd, tpl_batch_dd, custom_batch_tb, lang_batch_dd, model_batch_dd],
         outputs=[batch_results_md, batch_file_out, batch_status_md, pwd_group],
     )