CPM India AI Chatbot
A white-label AI chatbot platform that clients can run on their own servers.
- Client
- CPM India Sales and Marketing
- Through
- Techindika
- Role
- Full-stack developer
- Focus
- Frontend, in a team
- Timeline
- Jul 2025 to Aug 2025
- Code
- PRJ-02 · 2025
The problem
CPM India needed a chatbot they could resell to FMCG and BFSI clients under the client's brand. Existing products locked them into one LLM vendor, ran cloud-only, and could not keep one client's data away from another's.
None of the tools could answer from live database or API data without custom middleware written per client.
The build
We built a platform that deploys on-prem or in the cloud. Every workspace has its own knowledge base, agent persona and widget settings, and nothing leaks between them.
The RAG pipeline ingests documents, spreadsheets, SQL queries, REST endpoints and web pages into Azure Cognitive Search. Responses stream from Azure OpenAI, with Ollama as the on-prem option for clients who cannot send data out.
The public widget lives in an iframe so the client's CSS and scripts cannot break it. Authentication uses a public key exchanged for a short-lived token, with refresh handled before expiry so a session never dies mid-conversation.
Scope
- Workspace isolation per tenant
- License-key reseller model
- Custom agent roles and personas
- Docker and Kubernetes deployment
- RAG over documents, Excel, SQL and REST
- Scheduled data refresh
- Streaming responses
- Unanswered query tracking
- Two-factor authentication
- Role-based access control
- Token bucket rate limiting
- Iframe-isolated embeddable widget
Stack
- Next.js
- React
- TypeScript
- Azure OpenAI
- Claude API
- Ollama
- Playwright
- MongoDB
- PostgreSQL