Polish for judges: cold-start hint, example prompts, graceful errors, prewarm

- Cold-start note under the command bar so a 30-60s first request reads as
expected, not broken
- Clickable example prompts (gr.Examples) so a cold visitor reaches the magic
moment without staring at an empty box
- Graceful failure in on_command/on_brief: a model cold-start or timeout shows a
retry hint instead of a raw stack trace, and keeps the typed text in the box
- Best-effort startup prewarm thread so the Space self-warms both Modal models
- Expand README with explicit Build Small track/prize/badge write-up

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

README.md

@@ -88,4 +88,34 @@ EXTRACT_URL   ->  the Extractor.web URL  (MiniCPM3-4B)
 COACH_URL     ->  the Coach.web URL      (Nemotron-3-Nano-4B)
 ```
 
-Built for the build-small-hackathon. Models by OpenBMB and NVIDIA, GPUs by Modal.
+## Built for Build Small
+
+**Track — Backyard AI.** A practical, problem-solving tool for the daily work of a
+solo health coach: it removes the manual CRM upkeep so they can spend their time
+coaching, not bookkeeping.
+
+**How each prize is earned:**
+
+- **Modal** — both models run entirely on Modal. `modal_app.py` defines two GPU
+  services (L4), each warm-loaded and scaling to zero between calls; the cockpit is
+  a thin client that reaches them over HTTP. Modal is the whole runtime, not a side
+  call.
+- **OpenBMB (MiniCPM3-4B)** — the write path. One model does routing, structured
+  extraction, stage classification, and method harvesting, leaning on MiniCPM3's
+  strong format- and JSON-following.
+- **NVIDIA (Nemotron-3-Nano-4B)** — the read path. Compact agentic reasoning over
+  the coach's own data to answer questions and write grounded briefs.
+- **Tiny Titan** — every model the app depends on is ≤4B parameters.
+- **Best Agent** — an update flows through a multi-step pipeline: route → extract →
+  classify stage → log timeline → harvest reusable methods, then surfaces what needs
+  attention. Planning and tool use, not a single prompt.
+- **Off Brand** — a purpose-built CRM cockpit (command bar, pipeline board,
+  needs-attention panel, knowledgebase), not the default Gradio look.
+
+**Honest small-model fit.** Neither model does medicine. One turns natural language
+into a structured record; the other reasons only over text the coach supplied and is
+told never to introduce outside clinical claims. Every brief ends with "Not medical
+advice. Coach reviews before sending," and nothing is ever sent to a client
+automatically.
+
+Models by OpenBMB and NVIDIA, GPUs by Modal.

app.py

@@ -26,6 +26,7 @@ import json
 import os
 import re
 import sqlite3
+import threading
 
 import gradio as gr
 import requests
@@ -476,13 +477,25 @@ def _refresh(active):
 
 
 def on_command(text, active):
-    """The single command bar. Routes to update or query."""
+    """The single command bar. Routes to update or query, with graceful failure if a
+    model is cold or times out so a live demo never surfaces a raw stack trace."""
     text = (text or "").strip()
     if not text:
         out = "Type an update or a question."
         return (out, record_html(active), board_html(), needs_html(),
                 gr.update(), gr.update(choices=[], value=[]), [], "")
+    try:
+        return _handle_command(text, active)
+    except Exception:
+        out = ("⚠️ The model is waking up or the request timed out. The first request "
+               "after idle can take ~30–60s while the models load on Modal — give it a "
+               "moment and send again.")
+        # Keep the text in the box so the coach can just retry.
+        return (out, record_html(active), board_html(), needs_html(),
+                gr.update(), gr.update(choices=[], value=[]), [], text)
+
 
+def _handle_command(text, active):
     route = _post(EXTRACT_URL, {"op": "route", "text": text, "clients": list_clients()})
     intent = route.get("intent", "update")
     if text.endswith("?"):
@@ -549,11 +562,15 @@ def on_seed_kb(title, body):
 def on_brief(name, question):
     if not name:
         return "Pick a client first."
-    record = get_client(name)["record"]
-    kb = retrieve_kb(record)
-    out = _strip_think(_post(COACH_URL, {"op": "brief", "record": record, "kb": kb, "question": question})["brief"])
-    used = ", ".join(e["title"] for e in kb) if kb else "none yet"
-    return f"**Knowledgebase used:** {used}\n\n{out}"
+    try:
+        record = get_client(name)["record"]
+        kb = retrieve_kb(record)
+        out = _strip_think(_post(COACH_URL, {"op": "brief", "record": record, "kb": kb, "question": question})["brief"])
+        used = ", ".join(e["title"] for e in kb) if kb else "none yet"
+        return f"**Knowledgebase used:** {used}\n\n{out}"
+    except Exception:
+        return ("⚠️ The coach model is waking up or timed out. The first request after "
+                "idle can take ~30–60s on Modal — give it a moment and try again.")
 
 
 # ---------------------------------------------------------------------------
@@ -570,6 +587,7 @@ CSS = """
 #tag { color:var(--muted); margin-top:-.35rem; font-size:.9rem; }
 .muted { color:var(--muted); }
 #cmdout { background:var(--panel); border:1px solid var(--line); border-radius:12px; padding:6px 14px; min-height:2.2rem; }
+#coldhint { color:var(--muted); font-size:.78rem; margin:2px 2px 8px; }
 .record { background:var(--panel); border:1px solid var(--line); border-radius:14px; padding:16px 18px; }
 .rec-head { display:flex; align-items:center; gap:10px; }
 .rec-name { font-weight:800; font-size:1.15rem; }
@@ -599,7 +617,24 @@ ul.flags li { color:var(--warn); } ul.follow li { color:#9fd0ff; }
 .gr-button-primary { background:linear-gradient(90deg,#3fcfa0,#4f84ff) !important; border:none !important; }
 """
 
+def _prewarm():
+    """Fire one cheap request at each model on boot so the first real interaction is
+    warm, not a 30–60s cold start. Best-effort and off the request path: runs in a
+    daemon thread, and any failure (endpoint unset, still loading) is ignored."""
+    for url, payload in (
+        (EXTRACT_URL, {"op": "route", "text": "hello", "clients": []}),
+        (COACH_URL, {"op": "answer", "context": "{}", "question": "hello"}),
+    ):
+        if not url:
+            continue
+        try:
+            requests.post(url, json=payload, timeout=180)
+        except Exception:
+            pass
+
+
 db_init()
+threading.Thread(target=_prewarm, daemon=True).start()
 _initial = (list_clients() or [None])[0]
 
 with gr.Blocks(title="Spruce") as demo:
@@ -618,6 +653,20 @@ with gr.Blocks(title="Spruce") as demo:
             lines=2, scale=5,
         )
         send = gr.Button("Send", variant="primary", scale=1)
+    gr.Examples(
+        examples=[
+            ["Had a discovery call with Max, wants to drop 5kg before September and fix afternoon energy crashes. Booked his analysis call Thursday. I always have clients front-load protein at breakfast when energy dips."],
+            ["Dana reported her 3pm energy crash is gone after the protein change. Moving her into weeks 5-8."],
+            ["What's the update on Dana?"],
+            ["Who have I not contacted in a while?"],
+        ],
+        inputs=cmd,
+        label="Try one (click to load, then Send):",
+    )
+    gr.HTML(
+        "<div id='coldhint'>First request after idle can take ~30–60s while "
+        "MiniCPM3-4B and Nemotron-3-Nano-4B wake on Modal — then it's fast.</div>"
+    )
     cmd_out = gr.Markdown(elem_id="cmdout")
 
     with gr.Row():