The problem
Small and medium businesses get the same questions every day: What are your prices? Are you open on Friday? Where is my order? Answering them takes hours, and customers who message at night wait until morning.
The solution
Technov is a multi-tenant SaaS platform that gives every business its own AI customer support agent. A company uploads its documents, connects its WhatsApp number or website, and the agent starts answering straight away, using only that company's knowledge. When the AI can't help, or the customer asks for a person, the conversation is handed to a human agent in one click.
Key features
- AI that knows the business: PDF and DOCX documents are chunked, embedded and indexed in Pinecone, then reranked with Cohere before OpenAI writes the answer (retrieval-augmented generation).
- Omnichannel inbox: WhatsApp Business (Meta Embedded Signup) and a REST API for websites and apps, all in one live inbox with real-time updates over Socket.io.
- Human handoff: the AI escalates when needed, agents take over the live chat, and hand it back to the AI when done. Tickets keep the full history.
- Built for SaaS: isolated organizations, email invitations, API keys (create, rotate, revoke), subscription plans with limits and feature flags, and per-organization usage tracking.
- Analytics: separate dashboards for platform admins, company admins and staff, covering AI resolution rate, peak hours, channel mix and staff performance.
- Business tools: the AI can create orders and tickets straight from the chat, plus a searchable contacts book.
How it works
- Get invited: the platform admin invites the company; the owner completes onboarding and receives an API key.
- Train the AI: upload FAQs, price lists, policies and company information to the knowledge base.
- Connect channels: link a WhatsApp Business number, or call the API from a website.
- Go live: the AI answers customers 24/7 while the team watches, takes over and analyses from the dashboard.
For every message, the API checks the organization's plan and usage limits, detects the intent (greetings and handoff requests skip the AI entirely to save tokens), retrieves and reranks the relevant knowledge, and sends the grounded reply back to the channel.
Security
- JWT for the dashboard and per-organization API keys for integrations
- Passwords hashed with bcrypt; API keys encrypted at rest with AES-256-GCM
- Every query scoped to the caller's organization for tenant isolation
- WhatsApp webhooks verified with an HMAC signature
- Rate limits on the API, login, AI and uploads; Helmet headers and Zod validation everywhere
Tech stack
- Frontend: Next.js 16, React 19, TypeScript, Tailwind CSS 4, shadcn/ui, TanStack Query, React Hook Form, Zod, Recharts
- Backend: Node.js, Express 5, TypeScript, Mongoose, Socket.io
- AI: OpenAI, Hugging Face embeddings, Pinecone, Cohere rerank
- Data & integrations: MongoDB, WhatsApp Cloud API, Backblaze B2, Inngest, Nodemailer
- Quality: Jest test suite, GitHub Actions CI, Docker Compose