Smarter Architecture at a glance

  • Scalable serverless Kubernetes compute infrastructure that is largely self-maintaining.
  • Countermeasures for common Internet and web intrusion strategies including SQL injection, cross-site request forgeries, brute force password attacks, distributed denial of service, cross-site scripting, clickjacking, host header corruptions. Additionally, Smarter goes to great pains to minimize its attack surface, primarily by only opening ports 80 and 443 to the public.
  • Python-Django customer web dashboard application for developing plugin-based API’s deployed to custom domains.
  • LangChain managed LLM API requests. This provides a layer of abstraction between Smarter and underlying LLM vendor providers, and it also provides a simple means of standardizing Smarter customers’ API format.
  • React.js sandbox chat UI for prototyping pre-production APIs. Also works as a skinnable stand-in production UI if a customer want this.
  • Customer API logging architecture implemented with Django models, signals and Celery tasks.
  • Customer admin team management features.
  • Configurable use-based billing features based on API calls as well as plugin usage.

Smarter.sh Hosted Service

Customers who use the Smarter.sh SaaS service access this cloud-hosted service architecture.

Python Django

Most of Smarter is developed using Python’s Django web framework with the following noteworthy additions:

  • Django-rest-knox, used for creating secure, performant REST APIs.
  • Django Celery, a robust asynchronous compute layer using Celery, Redis, and Kubernetes which can be leveraged for scheduled tasks like automated reports as well as real-time compute-intensive functions.
  • Pydantic, for extending Django’s settings module to facilitate CI-CD friendly configuration data from multiple sources: environment variable, terraform, Kubernetes secrets, Github Actions secrets, etc.
  • Open-source Python libraries, including Pandas, NumPy, SciPy, and Levenshtein.
  • OpenAI as the default LLM integration.  Other LLMs can also be used.
  • LangChain to manage chatbot data flow.

ReactJS chat application

The chat app in the dashboard sandbox is written in ReactJS. The React app leverages Vite.js, @chatscope/chat-ui-kit-react, and react-pro-sidebar.

Webapp design features

  • robust, highly customizable chat features
  • A component model for implementing your own highly personalized OpenAI apps
  • Skinnable UI for each app
  • Includes default assets for each app
  • Small compact code base
  • Robust error handling for non-200 response codes from the custom REST API
  • Handles direct text input as well as file attachments
  • Info link to the OpenAI API official code sample
  • Build-deploy managed with Vite

Smarter REST API

Not to be confused with Smarter’s flagship product, customer-implemented custom REST API’s, Smarter additionally has its own chatbot management REST API, which is a Python Django project implementing Querium’s proprietary Plugin model, along with additional models for commercializing the service.

Chatbot management API end points