06-14, 13:30–15:00 (Europe/London), Salisbury
In this tutorial we will build a AI system to assist you in finding the best bar for you to go to in London - maybe even this evening after the PyData conference.
We will build a personalized recommender service that predicts the best venue for a drink for you based on your preferences (busy or less busy) and real-time Foot traffic Data.
We will then sprinkle some AI magic dust on the service by enabling you to ask the service questions, like - "can you recommend a bar within 15 minutes walking distance"?, using a LLM that has been fine-tuned to support function calling.
This tutorial may sound daunting, but we will only be really building 3 programs and a UI. We will use only Python, free serverless ML infrastructure, provided Hopsworks, and an API providing real-time footfall data.
Your ML system will consist of 3 ML pipelines:
- a feature pipeline to create the features for training and inference
- a training pipeline to take features and create a personalized ranking model
- an inference pipeline to find candidate bars and rank them based on your preferences
We will then build an open-source LLM-powered UI in Python so that you can interact with your system in natural language. The LLM will use the function calling paradigm to query the structured data in Hopsworks Feature Store.
Technologies
We will write all our programs in Python and use free serverless ML infrastructure, provided Hopsworks, and an open-source LLM.
After all that hard work, you’ll know the best place for you to go and get a beer.
Main repository;
https://github.com/MagicLex/workshop
You will need a free account on those two platforms;
https://app.hopsworks.ai
https://besttime.app/
Download the following libraries;
langchain-community==0.0.38
langchain-core==0.1.52
xgboost==2.0.3
transformers==4.38.2
protobuf==3.20.0
langchain==0.1.10
streamlit==1.31.1
sentencepiece==0.2.0
gradio==4.21.0
torch==2.3.1
pandas==2.1.4
hopsworks==3.7.0
seaborn==0.13.2
No previous knowledge expected
Lex Avstreikh is a machine learning strategist with a solid track record in enhancing operational systems and advancing ML infrastructure technologies. At Hopsworks, he is part of the team in charge of designing new innovative capabilites, leveraging his expertise in MLOps, to establish strong market positions against industry giants. Lex is also contributing to the use of F.A.I.R. principles in multiple research projects accross Europe.
Raymond Cunningham is a builder of distributed systems with 20+ years of experience in a number of different startups that he either co-founded or was one of the first employees. Currently, along with the Hopsworks engineering team, he is building a best in class MLOps platform and feature store to accelerate the delivery of cutting edge machine learning solutions.