Author: Twinkle Viswanathan Co-Authors: Vivek Kanna Jayaprakash, Panga Karthik, Rahul Kumar Goyal, Tejaswini R
Problem
In the manufacturing industry, workers often face complex tasks and machinery that require precise handling and understanding. Despite training, they may encounter unfamiliar situations, leading to time-consuming searches through printed manuals or delays waiting for supervisory assistance. Sometimes the workers might also feel too shy to question the trainer or authorities in case of doubts.
Proposed Solution
We propose deploying an advanced Automated Manuals Content Extractor application integrated with the company’s digital worker manuals. This application provides instant, reliable, and interactive assistance, enabling workers to query in natural language about any task or procedure.
How Do We Build It?
The required machine operation and training manuals are loaded to the Snowflake internal stage and Cortex’s Vector Embedding function to vectorize the chunks of the manuals(previously loaded in the internal stage) and store them in a datatype column in a Snowflake table along with the details of the manual and extracted chunk. Cortex’s VECTOR COSINE SIMILARITY function is used to find the most relevant chunks, and Cortex’s COMPLETE function to generate responses based on the user-provided model to implement a RAG-based LLM chatbot.
How Do We Use It?
In the sidebar(left pane) we have the option to choose
- The type of LLM model from below
- ‘mixtral-8x7b’
- ‘snowflake-arctic’
- ‘mistral-large’
- ‘llama3-8b’
- ‘llama3-70b’
- ‘reka-flash’
- ‘mistral-7b’
- ‘llama2-70b-chat’
- ‘Gemma-7b’
- Whether the previous question and response history is to be maintained and considered for the subsequent questions
- Whether we need the assistant to answer from the given set of operation and training manuals or if it can answer from the information available in the internet
- Checkbox to show the summary of the previous conversations
- A download button that enables the download of the previous conversations
In the main pane, we have the chat box to ask questions and the response is shown below:
Future Enhancements
Currently, the app only supports questions and responses through text, we could enhance the app to support images as well in the prompt and responses. Furthermore, information related to employee experience and skills could be added to enhance the quality of response and improve user experience.
