✏️ [TRIAGE] NOTEPAD for [BUG] #23: - CJ Worksheet - Model: Fine Tuning - Vector Storage | Prompts

[ai-avengers-team]

Dedicated Notepad for shared notes on #23 For @iPoetDev benefit.

Notes, WIP and recording what we need to do ... here. Updates #23 when realised.

See Private

View the Private notebook for full Triage

Code Flow

Explanation

• Start: Represents the beginning: environment is set up.
• Load Environment and Libraries: Involves loading environment variables and necessary libraries.
• Define Components:
  • This decision node branches into defining various components of the application.
• Initialize Chain:
  • Sets up the language model and vector store.
• Main Function:
  • Handles the main logic of the application, setting up the Streamlit interface and managing user interactions.
• User Input?:
  • A decision node that checks if there is user input.
• Get Augmented Query:
  • Prepares the query for the language model.
• Invoke QA Chain:
  • Calls the conversational retrieval chain to get a response.
• Display Response and Sources:
  • Shows the chatbot's response and any source documents.
• End: Marks the end of the flow

flowchart TD
    A[Start] --> B[Load Environment and Libraries]
    B --> B1[Import os]
    B1 --> B2[Import warnings]
    B2 --> B3[Import ChatGoogleGenerativeAI from langchain_google_genai]
    B3 --> B4[Import Chroma from langchain_chroma]
    B4 --> B5[Import ConversationalRetrievalChain from langchain.chains]
    B5 --> B6[Import SentenceTransformer from sentence_transformers]
    B6 --> B7[Import load_dotenv and find_dotenv from dotenv]
    B7 --> B8[Import Embeddings from langchain.embeddings.base]
    B8 --> B9[Import ConversationBufferMemory from langchain.memory]
    B9 --> B10[Import streamlit as st]
    B10 --> C{Define Components}
    C --> D[Define SentenceTransformerEmbeddings Class]
    C --> E[Define initialize_chain Function]
    E --> F[Initialize Chain with LLM and Vectorstore]
    F --> G[Define get_augmented_query Function]
    G --> H[Main Function]
    H --> I[Setup Streamlit Interface]
    I --> J{User Input?}
    J -->|Yes| K[Get Augmented Query]
    J -->|No| L[Wait for Input]
    K --> M[Invoke QA Chain]
    M --> N[Display Response and Sources]
    N --> O[End]

[ai-avengers-team]

Request to GCP Mentor

Early POC implementation for our chatbot was LangChain and ChromaDB

• ChromaDB for local development requires filestorage / blob / file path locally.

We have an ADR: Architectural choice:

• Continue to use Chroma (with LangChain access to Google Gemini 1.5) for dev and deploy
• Switch to GCP direct equivalents for ChromaDB
• Redirect to new path.

How would you like the mentor to help?

Q1. Please help with the ADR for tech choice and compatibility? Q2: Please advise if Google has a Vector database offering or if existing have support for Vector databases/embeddings? Q3: If we use ChromaDB in GCP/Google Cloud, how can we support its functions and needing to write to disk, when we deploy our app?