Interactive Knowledge Agent

Final course project (lab 9) for Laboratory Course of Machine Intelligence, Peking University, 2025 Fall.

An LLM-based interactive agent that models a learner’s understanding by adaptively probing topics from lecture materials through observation, planning, and reflection.

Implemented as a CLI tool that extracts topics from a lecture PDF, quizzes the user, reflects on answers, and tracks knowledge state. Supports topic number limiting and verbose mode with token usage.

Setup

1. Python 3.12
2. Install dependency:

pip install -r interactive_knowledge_agent/requirements.txt

3. Set OpenAI key or call your API key later:

export OPENAI_API_KEY=<your_openai_api_key>

Usage

From repo root:

python interactive_knowledge_agent/main.py \
  --pdf interactive_knowledge_agent/pdf_materials/example.pdf \
  --prompt-dir interactive_knowledge_agent/prompts \
  --output-dir interactive_knowledge_agent/outputs \
  --model gpt-4o \
  --max-topics 3 \
  --verbose \
  --api-key "$OPENAI_API_KEY"

Flags

• --pdf (required): path to lecture PDF.
• --prompt-dir: prompt templates directory (default prompts).
• --output-dir: where session JSONs are saved (default outputs).
• --model: OpenAI chat model name (default gpt-4o).
• --max-topics: keep only the first N topics from analysis.
• --verbose: print topics, per-turn reflections, knowledge updates, and token usage.
• --api-key: optional; otherwise uses OPENAI_API_KEY env.