Overview

Demo Link:

TBD

VoiceChat

• Enables user to have a conversation with an LLM (Voice in, Voice out)
  • Voice input is working, Voice output is currently being constructed
• Enables user to select a narrator that will act according to the description. (The TTS models that will be invoked are part of the next build cycle: AI)
  • Pirate
  • Scotsman
  • More

TextChat

• Enables user to interact with an LLM via prompt
• Auto-Complete (part of frontend build cycle)

Technologies Applied

LLM

• ChatGPT

Backend Server

• FastAPI - Text-To-Text and Speech-To-Speech Flows (Next Build Cycle)
• Django - User creation, authentication, and authorization

Languages

• React
• Typescript
• Python
• CSS: Tailwind

Containerization

• Docker (Next build cycle)

Streaming flow of audio input

• Websockets

Databases

• PostGreSQL (User/Auth - Next Build Cycle)
• VectorDB (Domain specific knowlegdge - RAG architecture)

Future Planning

Conversations

• User will be able to select any of their previous conversations (limit of X)
• When selected the convo will populate into the history window above the TextPrompt

Documents

• User will be able to upload a doc that will either:
  • passed in as a one-off for the text or voice-chat prompt
  • stored permanently as a doc that is referenced when they query the LLM

User Account Creation and Auth

• A new service will be constructed in Django to enable noobs to:
  • Register an account
  • Log in
  • Delete the account
  • Auth for use of LLM

Clustering

• EKS / Kubernetes / Terraform

LLM To SLM

Swap out ChatGPT for an SLM like Llama3 that is running on our own server.

Mobile Versions

• iOS
• Android