UI
- today's date in the masthead ("Account for <date>") + every page's section-head,
with a festival-countdown chip; locale-independent EN/हिं names in i18n.py
- empty Festival card now surfaces the next upcoming festival (proactive.festival_nudge)
- Talk page starter chips; a visible "how Saathi answered" tool-call trace under each
reply (Best Agent); Receive page reordered + gated hint; Khata reminder-pill hover
cue; composer caption
Reliability / latency
- LLM_REQUEST_TIMEOUT default 120 -> 180 (survive a Modal cold start)
- scripts/prewarm.py: warm LLM + STT + TTS before a demo
- modal_app.py: pre-load Whisper + Veena at container startup (fixes the voice
"responds too late" lag — no more lazy model load on the first /stt and /tts)
Docs / content
- fix stale final_report.md (MMS -> Veena; "fully local" -> open models on Modal)
- README: Hosting & deployment, Tracks & badges, Evaluation sections
- Field-Notes blog draft + social drafts; scripts/export_trace.py (Sharing-is…
