Generative AI for Finance

Unit 1: Working with APIs

• REST API basics
• Working with OpenAI’s API offerings and python client
• Working with Huggingface’s python client
• Working with Weaviate, a vector database

Unit 2: Build your own semantic search engine

• Preparing data
• Choosing and using a vectorize
• Defining a schema and importing data
• Querying and controlling output

Unit 3: Connecting semantic search to LLMs

• Retrieval augmented generation
• Open domain question answering
• Summarization and main idea extraction
• Sentiment analysis

Unit 4: Roll your own Bloomberg GPT

• Question answering
• Summarization
• Sentiment analysis

Unit 5: Generating synthetic data

• What is synthetic data and when should you use it?
• Synthetic generation of private data
• Stress testing on synthetic macroeconomic scenarios
• Synthetic data generation of macroeconomic scenarios with GPT-3
  Developed by NVIDIA – Yi Dong, Emanuel Scoullos

Unit 6: Using language models in your workflow

• Chain of thought reasoning to explain output
• Self debate to explore latent knowledge
• Using LLMs to write SQL queries
• Using LLMs to aid in data analysis

Datasets we will use:

Resources we will reference:

• https://developer.nvidia.com/blog/generating-synthetic-data-with-transformers-a-solution-for-enterprise-data-challenges/
• https://arxiv.org/abs/2106.09685
• https://platform.openai.com/docs/introduction/overview