Team Y

Blog

Home

Kamil Janeczek

June 18, 2024

AI assistant

Backend

GenAI JavaScript

LLM

RAG

Objective

Create personal AI powered assistant that can help with day to day task improve my performance

Technologies

Fastify

Node.js

JavaScript

TypeScript

OAuth2

Google API

Prisma

Zod

langachain

Vecel- /blob

Vercel- /kv

puppeteer

Social share

Challenge

Team Y application is a backend for AI assistant. It is fronted agnostic with user experiance in mind, only just a few of endpoint when interacting with assistant. Main features supported:

session management
security
conversation with bot
storage of conversations
long and short term memory
integration with Google service (calendar, tasks, gmail)
advanced RAG techniques see more in this post
ingestion of documents
voice transcrition

This application is built using TypeScript. It uses Node.js for runtime. The application uses Fastify, a fast and low overhead web framework for Node.js, for handling HTTP requests. The application uses Prisma, an open-source database toolkit, for database operations, dotenv for managing environment variables. Vercel/blob is used for file storage and Vercel/kv for Redis like database.

Here is a specyfication of API:

User

/register POST
/session POST
/login POST
/logout POST

Conversation

/ask POST
/snippets GET

Action

/transcribe POST
/transcribeAndAsk POST
/uploadfile
/uploadrecording

Chat

/create POST
/:chatId GET
/getchats/:botId GET

User needs to login to gain access to application. In response service returns auth token that needs to be attached to each and every requsest to the server. For browsers, token is automatically saved to cookies. If needed new account can be created.

Create chat

After sucessful login user choose we creat a chat in order to take advanatage of conversational capabilities of bot

Conversation with assistant

Using context of chat we send messages towards bot. API is simple and unified. One main endpoint /ask to communicate with bot.

Releated Projects