Multi-Channel AI
Advanced AI Automation Platform
UAT dry-run 11/11
WhatsApp, webchat, email, and voice on one FastAPI + n8n brain — RAG, CRM factory, JWT ops dashboard, booking, follow-ups, and handoff.
M0–M10 complete (~14 weeks) · API 1.0.0 · Architecture v1.1
The problem
Sales and support span multiple channels with no shared brain, weak grounding, and silent failures.
Fragmented channels
WhatsApp, web, email, and voice each had half-built bots with no shared brain or CRM sync.
Ungrounded LLM answers
Plain chatbots invent policy and pricing; teams lose trust.
Manual lead loops
Qualify → score → book → follow-up still lived in spreadsheets and inboxes.
Silent automation failures
Workflows died without alerts; duplicates reprocessed webhooks.
Weak human handoff
Agents got raw transcripts with no AI pause/resume or structured package.
The solution
FastAPI owns AI, RAG, and domain logic; n8n orchestrates thin side-effect workflows only — no god-workflow.
A Dockerized multi-channel AI engagement platform for sales, support, lead qualification, booking, follow-ups, and ops — not a FAQ chatbot and not a single linear n8n flow. Every inbound event is classified, routed to a specialized agent, grounded in a private KB when needed, persisted to CRM, and resolved by AI or a human with full context.
Orchestrator + specialized agents
Sales, support, qualify, booking, RAG, escalation — routed by intent and guardrails.
Qdrant RAG with confidence gates
Citations on grounded answers; escalate instead of hallucinate.
CRM factory
HubSpot primary + GoHighLevel adapter; dry-run mock IDs when tokens empty.
Production reliability
Redis idempotency, booking locks, DLQ, ops notify, Prometheus/Grafana.
JWT RBAC ops dashboard
React SPA with admin / agent / viewer — viewer cannot mutate routing.
Channels → Caddy/FastAPI webhooks → Orchestrator → Agents + RAG → Postgres / Qdrant / Redis → n8n → CRM / Calendar / Stripe / Notify
Before → After
Architecture
Delivery
Milestone narrative from kickoff through closeout — all gates met.
| ID | Theme | Unlocked |
|---|---|---|
| M0–M1 | Architecture + skeleton | Compose stack, schema, Redis, FastAPI, health, n8n error WF |
| M2–M3 | WA + RAG + CRM + booking | Orchestrator, Qdrant KB, HubSpot, Calendar locks, lead journey |
| M4–M5 | Handoff + follow-ups | HITL packages, DLQ, sequences, KPI API, notify matrix |
| M6–M7 | Email + voice | Attachment intelligence, voice dry-run, Phase 2 gate |
| M8–M9 | Dashboard + scale | React SPA, RBAC, GHL, Stripe, Prometheus/Grafana, webchat |
| M10 | Closeout | Docs, UAT, KT pack, deliverables tracker |
Proof
Verified on local Docker dry-run — no invented client metrics.
UAT dry-run
11/11 Pass
Chat, lead, booking, handoff, email, voice, RBAC, webchat, payment, KB re-ingest
Regression pack
OK 3/3
phase2 + m8 + m9 via test_m10_pack.py
Viewer RBAC
403
PUT /admin/routing denied for viewer role
PRD deliverables
24/24
Done with documented waivers
Stack
- FastAPI
- n8n
- React
- Postgres
- Redis
- Qdrant
- OpenAI
- HubSpot / GHL
- Stripe
- Prometheus / Grafana
- Caddy
- Docker Compose
Want a system like this?
We build production AI ops platforms — multi-agent orchestration, RAG, CRM, and reliable automation — not one-off chatbot demos.