Awesome
Autonomous HR Chatbot built using ChatGPT, LangChain, Pinecone and Streamlit
Companion Reading: Creating a (mostly) Autonomous HR Assistant with ChatGPT and LangChain’s Agents and Tools
TL;DR/Description
This is a prototype enterprise application - an autonomous agent that is able to answer HR queries using the tools it has on hand. It was made using LangChain's agents and tools modules, using Pinecone as vector database and powered by ChatGPT or gpt-3.5-turbo. The front-end is Streamlit using the streamlit_chat component.
Tools currently assigned (with more on the way):
- Timekeeping Policies - A ChatGPT generated sample HR policy document. Embeddings were created for this doc using OpenAI’s text-embedding-ada-002 model and stored in a Pinecone index.
- Employee Data - A csv file containing dummy employee data (e.g. name, supervisor, # of leaves etc). It's loaded as a pandas dataframe and manipulated by the LLM using LangChain's PythonAstREPLTool
- Calculator - this is LangChain's calculator chain module, LLMMathChain
Sample Chat
Sample Tool Use
Instructions
I made this prototype using Azure deployments as my company is an Azure customer.
I created a backend file called hr_agent_backend_local.py
for those that does not want to use Azure.
This is does not use any Azure components - the API is from platform.openai.com, the csv file is stored locally(i.e. on your own computer)
How to use this repo
- Install python 3.10. Windows, Mac
- Clone the repo to a local directory.
- Navigate to the local directory and run this command in your terminal to install all prerequisite modules -
pip install -r requirements.txt
- Input your own API keys in the
hr_agent_backend_local.py
file (orhr_agent_backend_azure.py
if you want to use the azure version; just uncomment it in the frontend.py file) - Run
streamlit run hr_agent_frontent.py
in your terminal
Storing Embeddings in Pinecone
- Create a Pinecone account in pinecone.io - there is a free tier. Take note of the Pinecone API and environment values.
- Run the notebook 'store_embeddings_in_pinecone.ipynb'. Replace the Pinecone and OpenAI API keys (for the embedding model) with your own.
Tech Stack
Azure OpenAI Service - the OpenAI service offering for Azure customers.
LangChain - development frame work for building apps around LLMs.
Pinecone - the vector database for storing the embeddings.
Streamlit - used for the front end. Lightweight framework for deploying python web apps.
Azure Data Lake - for landing the employee data csv files. Any other cloud storage should work just as well (blob, S3 etc).
Azure Data Factory - used to create the data pipeline.
SAP HCM - the source system for employee data.
Video Demo
Roadmap
Currently working on adding the following tools using OpenAI's function calling feature:
- Currency Exchange Rate tool - this tool will have access to the internet to check the current FX rate. Sample HR Use Case: a contractor paid in USD can ask how much he will be paid in his/her local currency - e.g. 'How much is my salary this month in PHP?
- Tax Explainer - the employee can ask how his/her tax (and other deductions) are computed for the payroll period based on tax rates and statutory deduction tables e.g. taxable gross, social security deductions etc. The chatbot will illustrate how the tax/deduction was computed based on the user's own payroll data/values. Idea stolen (with permission) from Jem Rodil :)
Other suggestions welcome. ☺️
Just open a new topic in the discussions
section.
Author
Stephen Bonifacio
Feel free to connect with me on:
Linkedin: https://www.linkedin.com/in/stephenbonifacio/
Twitter: https://twitter.com/Stepanogil