Project Lifecycle Documentation

[akiwarheit]

Scope: Define the use case

To create a chatbot that provides property recommendations to users.

Select: Choose an existing model or pretrain your own

A couple of options here:

1. FLAN T5 - We can use this as a baseline (course also has code references so we can easily use that)

Adapt & Align: Prompt Engineering

Sure, but we will most likely be focusing on fine-tuning.

Adapt & Align: Fine-tuning

Fine-tuning the model is similar to "training" the model in traditional supervised learning approaches

This involves formatting the training data as follows:

or:

Prompt: Which county should I look for properties if I earn $100k a year?
Answer: {county_1}, {county_2}, {county_3} etc2.

Align with human feedback

Maybe ask friends & family?

Evaluate

ROUGE, BLEU

Application Integration

Build the actual chatbot interface

We can either create a Streamlit dashboard, ezpz, or flex our web development skills.