Intership/Master Thesis Build fast local AI apps with PostgresML

PostgresML is a PostgreSQL extension, which inverts the current paradigm of continuously moving the data to the Large Language Models (LLM). Instead, it moves the models to the (local) database. PostgresML allows you to take advantage of the fundamental relationship between data and models and thus allows for local access to the latest open-source ML & AI algorithms, eg from Huggingface inside the database with a SQL API. It also provides native language SDKs to access the models as well as support for GPU acceleration.
Local Named Entity Recognition (NER) models, eg openNLP have been around for some time and are easy to integrate into a local application. With the emerge of LLMs and their ability to provide NER functionality, people started to access these models via HTTP requests, eg the chat completions endpoint of GPT-4. However, the access via single HTTP requests is a performance bottleneck. Besides that, company employees are not supposed to expose confidential data to public services. PostgresML is here to change this. 
In this Internship, we want to gain experience with PostgresML and put it in context of other model serving frameworks. This Internship should also lay the groundwork for follow up projects in the realm of leveraging the potential of open-source LLMs run on-premise.

• Learn the features of PostgresML and gain experience with running it on a local machine.
• Develop a fast local NLP/NER example application. NER involves identifying and classifying entities in text into predefined categories such as names of persons, organizations, locations, dates, monetary values as well as domain specific terms, eg medical terms. 
• Evaluate the best matching open source local LLM for NER and research possibilities to further train this model. 
• Compare PostgresML to other model serving frameworks, eg BentoML, mosec or SparkNLP. 
• Assess the potential for other local NLP uses cases at ELCA and the means needed to implement them, eg embeddings and fine tuning.
• Present the results in an ELCA Brownbag session.

• A dynamic work and collaborative environment with a highly motivated multi-cultural and international sites team
• Personal development through training and coaching
• A culture of collaboration across all disciplines
• The chance to make a difference in peoples’ life by building innovative solutions
• High innovation and research backed up by collaboration with universities like EPFL
• Various internal coding events (Hackathon, Brownbags), see our  technical blog
• Monthly After-Works organized per locations
• Good work-life balance (2 days per week from home)

• Holding a technical bachelors degree and/or doing your masters
• Basic understanding of ML and LLMs
• Basic understanding of SQL, Java, Python 
• Interest in application architectures
• Interest in comparing and applying different development approaches