Installation

• Create a Python virtual environment. For example using venv:

python -m venv .llm
source .llm/bin/activate

• Install the requirements:

pip install -r requirements.txt

• Download the .env file and replace the OPENAI API key with your own if so choose: https://drive.google.com/file/d/14_6HYo3S4nQW9PNqjRbHEZG5-EPyNTWR/view?usp=sharing
• Go to the appropriate project directory, for example HED tagging bot:

cd hed-bot/hed-tagging-bot

• Run the bot in the desired port, for example 55000:

chainlit run --port 55000 app.py -w

Tech Stack

• Python: Main programming language.
• Chainlit: Conversational AI web UI framework.
• LangChain & LangGraph: LLM orchestration, prompt management, and agent workflows.
• OpenAI API: For LLM completions (via langchain-openai).
• HED Python Library: For Hierarchical Event Descriptor (HED) string validation and manipulation.
• BeautifulSoup & Requests: For web scraping and XML parsing (fetching/parsing HED vocabularies).
• dotenv: For environment variable management.

Codebase Overview

• The main project is organized under hed-bot/, with subfolders for each bot:
  • hed-tagging-bot: Translates event descriptions to HED annotations and validates them.
  • hed-doc-bot, hed-rag-tagging-bot: Other specialized bots, each with a similar structure.
• Each bot contains:
  • app.py: Main application logic, including Chainlit event handlers and LLM agent setup.
  • .chainlit/: Chainlit configuration (e.g., config.toml for UI and feature settings).
  • chainlit.md: Optional welcome screen/documentation for the bot UI.
  • (Some bots) Data files or additional resources (e.g., HEDLatest-terms).
• Key Components in hed-tagging-bot/app.py:
  • Loads environment variables and sets up the OpenAI LLM via LangChain.
  • Defines a tool for HED string validation using the HED Python library.
  • Fetches HED vocabulary from a local file or the official HED schema repository.
  • Sets up a Chainlit chat interface, where users can input event descriptions and receive HED annotations, with real-time validation feedback.
  • Handles image uploads and integrates them into the LLM prompt if provided.

Extending or Transferring

• To add new bots, replicate the structure of an existing bot directory.
• To change the LLM or prompt logic, modify the app.py in the relevant bot.
• To update the HED schema or vocab, replace the HEDLatest-terms file or adjust the fetching logic.
• Chainlit configuration and UI can be customized via .chainlit/config.toml and chainlit.md.