FinUA @ Junction 2024: Using Tech for Empathy and Connection (1st Session)

​Overview

FinUA embraces a holistic approach to supporting Ukrainians in Finland by integrating legal assistance, emotional support, and social adaptation—all through empathy and technology.

One important task FinUA uses to streamline operations is tagging user messages. This helps in organizing, prioritizing and distributing them to experts who can help with them. Currently, this is a manual process.

Recently, autoregressive Large Language Models have emerged as flexible tools for quickly prototyping NLP applications. Their combined language understanding, in-context learning and open-ended text generation capabilities make them extremely flexible tools for this purpose. One of the main ways to influence the model output is prompt engineering.

This can include writing instructions to the model, giving examples ("few-shot learning"), as well as more esoteric techniques that condition the model's outputs (e.g. offering a tip).

What You’ll Gain

In this workshop, we will use an autoregressive Large Language Model to tag messages, by calling the model from Python code. The goal is not to create a perfect tagging system, but to get a feel for how to develop with LLMs and what kinds of limitations one might encounter.

We will also share practical insights from FinUA’s journey since 2022, highlighting key technological solutions and tools that have transformed our operations.

We’ll cover:

• ​How AI optimizes volunteer work and problem-solving

• ​The concept of empathetic AI for better understanding user needs

• ​Strategies for effective matchmaking between volunteers and service seekers

​Keywords LLMs, legal support, immigration, integration, optimization, empathy

By focusing on actionable insights, we aim to provide participants with valuable tools and techniques they can apply in their own projects.

NB: If you are already an expert in NLP but are tired of LLMs, you are also free to use any other model or technique for this task. But for the sake of time, we recommend sticking to the LLM.