Beschreibung
You should set up an AI-project that improves the recruitment process by suggesting matches between candidates and jobprofiles.By now there is a prototype that calculates embeddings on candidates’ documents and jobprofiles and matches them as well as an ML-model that uses tabulardata to calculate matches between candidates and jobs.
The system ist set up as a local flask application with a REST-API for interaction running inside a docker container.
There are about 500 training items (matches of candidates and jobs).
What is missing:
The suggestions from the two models are not combined.
The two models deliver very different results
What you are expected to do:
rewrite application to use langchain instead of haystack
Analyze the ML-data statistically and clean the data.
Build a data-preprocessing-pipeline
combine the results from the two models
develop a different attempt and use text-embeddings as well as tabulardata in one individual model.
set up a training pipeline
configure the training pipeline for retraining the model in future
improve speed of server-backend by handling asynchronous server-calls
e.g. change to fastapi server backend
create api-documentation with swagger