NovaChat
v0.12.0 · Production · MVP lengkap

Chatbot klinik yang
belajar sendiri tiap minggu.

Platform untuk membangun chatbot klinik berbasis knowledge — cepat, aman, scalable. Tanpa LLM saat runtime, dengan kemampuan belajar mandiri setiap minggu.

Daftar Klinik Coba Widget Live

Fitur Lengkap

Semua yang dibutuhkan klinik untuk menjalankan chatbot 24/7 — dari builder admin sampai widget embed siap-pasang di website manapun.

⚡

Runtime < 2 detik, no LLM

Keyword pre-filter + FAISS embedding (multilingual MiniLM lokal) + rule engine. Zero hallucination, deterministik.

↺

Self-improving mingguan

Setiap Minggu malam, sistem ekstrak pola dari unresolved chat → Claude parse → knowledge baru. Senin pagi notif WA admin.

↑

Upload dokumen → knowledge

PDF, DOCX, TXT — parse async via Claude. Knowledge JSON terstruktur otomatis muncul di editor.

↗

Eskalasi cerdas

Keyword (komplain, manusia), loop unresolved, score-low, atau manual button → redirect ke WhatsApp admin dengan ringkasan chat.

⌘

Embed di web manapun

Satu <script> tag — widget chat siap pakai. Configurable warna, judul, posisi. Atau panggil API langsung.

⊞

Multi-tenant terisolasi

Setiap klinik punya knowledge, FAISS index, chat log, dan API key sendiri. JWT-scoped, zero cross-clinic access.

◐

Dashboard analytics

Chat per hari, top knowledge, top unresolved, breakdown status, eskalasi rate. Untuk admin klinik & super admin.

⌫

WhatsApp Cloud native

Webhook Meta WA Cloud API: incoming text → process → reply. Tiap klinik daftar WABA sendiri (model platform fee).

Status Sistem

API Status
Mengecek…
Tag Aktif
v0.12.0
embed-widget-deploy
Engine
FAISS + Rule
multilingual MiniLM
Komponen
5
API · Worker · Beat · Admin · Widget

Progres Build

Roadmap selesai — 12 milestone backend + frontend + deploy. Sumber: github.com/yanuarimansantosa/novachat (private).

✓v0.1.0Scaffolding — FastAPI + config + DB sessiondone
✓v0.2.0DB Schema — 8 tabel + Alembicdone
✓v0.3.0Auth — JWT + login/register/refreshdone
✓v0.4.0Knowledge CRUD — versioning + soft deletedone
✓v0.5.0Chat Runtime keyword — public /api/chatdone
✓v0.5.1Prod Preview — deploy + landing visualdone
✓v0.6.0FAISS embedding — multilingual MiniLMdone
✓v0.7.0Escalation — keyword/loop/manual/score-lowdone
✓v0.8.0Upload Parsing — Celery + Anthropicdone
✓v0.9.0Dashboard + History endpointsdone
✓v0.10.0WhatsApp Cloud + Weekly learning loopdone
✓v0.11.0Admin Builder SPA — full CRUD UIdone
✓v0.12.0Embed Widget + Final Deploydone

Arsitektur

Pemisahan core deterministik (rule-based + RAG, no LLM, < 2 detik) dari AI parser (Claude, async, untuk upload + weekly learning loop).

Client
Widget JS embed (widget.js · 8 KB) · WhatsApp Cloud · Admin Panel
↓
API Gateway — FastAPI (Uvicorn @ :8011)
/api/auth/* · /api/clinic/me · /api/knowledge/* · /api/chat · /api/jobs/* · /api/dashboard/* · /api/webhook/whatsapp
↓
Core Services (NO LLM at runtime)
Keyword pre-filter → FAISS vector search → Rule engine (confidence floor) → Template → Escalation
↓
Storage
PostgreSQL 16 (multi-tenant via clinic_id) · FAISS index per klinik · Redis (Celery broker)
↓
Async Worker (Celery + Beat)
novachat-worker: parse_upload, parse_weekly_learning → Claude · novachat-beat: weekly fanout (Sun 00:00)