🏠 Home > AI > AI Agent > Education Assistant > Class > Step 3

Step 3: Building the Brain (Agent Logic)

Now that we have memory, we need logic. The “Brain” needs to do three main things:

1. Understand Commands: Translate human language into database actions.
2. Monitor the World: Proactively check for problems.
3. Teach: Actively tutor students who need help.

1. Parsing Language (The “Ears”)

Teachers speak in natural language: “Schedule a review session for Friday.” The code needs strict instructions: INSERT INTO schedule VALUES ('2023-10-27').

The Hybrid Approach

We use a tiered system for “thinking”:

1. Cloud AI (Gemini/OpenAI): Best for complex reasoning.
2. Local AI (Ollama): Great for privacy and free usage. (Model: llama3 or olmo)
3. Regex Fallback: Simple pattern matching if AI fails.

# Pseudo-code logic
if has_gemini: call_gemini()
elif has_ollama: call_local_ollama()
else: use_regex()

2. The Daily Monitor (The “Conscience”)

This is what makes it an Agent and not just a chatbot. It acts without being asked.

Logic Flow:

1. Wake up (e.g., every morning).
2. Check Schedule: Is there class tomorrow? -> Queue Notification.
3. Equity Check: Does anyone have the needs_help tag?
   • Old Way: Send a static PDF.
   • New AI Way: Initiate a Tutoring Session.

3. The AI Tutor Loop

Instead of just flagging a student, the Agent enters an interactive loop:

1. Start: Agent generates a concept question.
2. Wait: Student responds (handled via process_student_message).
3. Evaluate: Agent (Ollama/Gemini) checks the answer.
   • Correct: “Great job!” (End Session)
   • Incorrect: Explain concept -> Ask new question -> Repeat.

4. Educational Goal

By building this, you learn:

• Parsing: Converting unstructured text (human) to structured data (machine).
• Local LLMs: Running AI models like Llama 3 locally using Ollama.
• Stateful Interactions: Managing long-running “Tutoring Sessions” in a database.