How We Built a Job Recommender SaaS with Deep Learning to Disrupt the Job Market

Discover how a team of Machine Learning engineers built a job recommender SaaS using Deep Learning to revolutionize the job market in this 45-minute conference talk. Learn about JobNet, a deep neural network that processes résumés and job descriptions in multiple languages, creating embeddings in a shared space for efficient job seeker and job matching. Explore the modern ML stack utilizing Dask, Sklearn, and TensorFlow, and gain insights into the cloud deployment process using a Continuous Integration pipeline with CircleCI, Terraform, Docker, and AWS ECS. Delve into the architecture of JobNet, including word and document embeddings, CNN techniques, and the Dask-orchestrated task graph. Understand the challenges of measuring recommendation quality and the benefits of Deep Learning in creating useful representations for variable-length inputs. Gain valuable knowledge on automating deployment and ensuring reproducible infrastructure and software in this comprehensive overview of building a disruptive job recommender system.

Intro
We are a team of Machine Learning engineers
Step 1/2: Use Deep Learning to learn embeddings
Step 2/2: Use embeddings to recommend jobs
How do you measure the quality of a list of jobs?
Evaluation measure for implicit missing feedback
Why Deep Learning?
Why use Deep Learning? 2 Useful representations
Why use Deep Learning? 3 Variable length input
Word embeddings learn to capture semantics
JobNet is a cascade of useful representations
Document embeddings with CNN52
JobNet's architecture
Dask orchestrates the full task graph
Automating deployment with CI/CD
Reproducible infrastructure & software