Teaching Assistant Agent

An adaptive teaching agent that explains any concept at your exact comprehension level, from absolute beginner to domain expert. Works across STEM, humanities, creative arts, and professional domains with multi-platform integration for learning management systems and note-taking tools.

Supported Platforms & Integrations

Platform	Integration Type	Features
Notion	API / Export	Saves lesson notes, creates study databases, tracks learning progress
Anki	Export	Generates spaced-repetition flashcard decks from completed lessons
Obsidian	Markdown Export	Creates linked knowledge graphs from explained concepts
Google Docs	Export	Formats explanations as shareable study documents
Quizlet	Export	Builds study sets with terms, definitions, and practice questions
Khan Academy	Reference	Cross-references explanations with KA curriculum paths

When to Use This Agent

• Use this when you need a complex topic broken down into digestible pieces with real-world analogies

• Use this when preparing for an exam and need practice problems with step-by-step solutions

• Use this when switching careers and need to rapidly learn a new domain from scratch

• Use this when you understand the basics but have specific gaps in advanced subtopics

• Use this when you want Socratic questioning to deepen understanding rather than passive reading

• Use this when you need the same concept explained multiple ways until it clicks

• Consider alternatives when you need certified course completion or official credentials

• Consider alternatives when you need real-time lab work or hands-on physical practice

• Consider alternatives when you need peer discussion and collaborative learning environments

Quick Start

Copy
# .claude/agents/teaching-assistant-agent.yml
name: Teaching Assistant Agent
model: opus
trigger: "teach me"
context:
  learner_level: intermediate
  preferred_style: visual-analogies
  domain_focus: general
  question_mode: socratic

First Run Example:

User: Teach me how neural networks learn through backpropagation

Agent: Let me gauge your current level first.

**Quick Assessment:** Can you tell me what you already know about:
1. What a neural network is conceptually?
2. How a function's derivative relates to its slope?
3. The chain rule in calculus?

[Based on your answers, I'll calibrate the explanation depth]

Expected Output:

## Backpropagation Explained (Intermediate Level)

**Analogy:** Imagine a factory assembly line where each station 
adjusts a product. When quality control at the end finds a defect,
they send feedback BACKWARD through each station telling them 
exactly how much to adjust.

**Core Mechanism:**
1. Forward Pass — Input flows through layers, producing output
2. Loss Calculation — Compare output vs expected (the "defect")
3. Backward Pass — Compute each weight's contribution to error
4. Update — Adjust weights proportionally using learning rate

**Practice Problem:** Given a 2-layer network with weights 
w1=0.5, w2=0.8 and input x=1.0, target=0.3...
[step-by-step solution follows]

Advanced Configuration

Platform-Specific Setup:

Copy
# Anki flashcard generation
anki:
  deck_name: "{{domain}}_concepts"
  card_type: cloze
  include_diagrams: true
  difficulty_tags: true

# Obsidian knowledge graph
obsidian:
  vault_path: ~/Documents/StudyVault
  link_related: true
  tag_prefix: "learn/"
  create_moc: true

Parameter	Type	Default	Description
learner_level	string	intermediate	beginner, intermediate, advanced, expert
preferred_style	string	mixed	visual-analogies, mathematical, narrative, code-examples
domain_focus	string	general	cs, math, physics, biology, history, economics, etc.
question_mode	string	socratic	socratic, direct, quiz, practice-problems
explanation_depth	int	3	1 (brief) to 5 (exhaustive) layers of detail
analogy_source	string	everyday	everyday, technical, sports, cooking, gaming
practice_count	int	3	Number of practice problems per concept
flashcard_gen	bool	true	Auto-generate spaced repetition cards
prerequisite_check	bool	true	Verify prerequisite knowledge before advancing
output_format	string	markdown	markdown, anki-csv, quizlet-tsv, notion-blocks
language	string	en	Explanation language (en, es, fr, de, ja, zh)

Core Concepts

Concept	Description
Adaptive Leveling	Dynamically adjusts explanation complexity based on comprehension signals
Socratic Method	Uses guided questions to lead learners to discover answers themselves
Scaffolding	Builds on confirmed knowledge, adding conceptual layers incrementally
Multi-Modal Explanation	Same concept delivered via analogy, math, code, and visual description
Spaced Repetition	Generates review prompts at optimal intervals for long-term retention

+------------------+     +------------------+     +------------------+
|  Learner Input   |---->|  Level Detector  |---->|  Style Selector  |
|  (question/topic)|     |  (assess prior   |     |  (analogy, math, |
|                  |     |   knowledge)     |     |   code, visual)  |
+------------------+     +------------------+     +------------------+
                                                          |
                                                          v
+------------------+     +------------------+     +------------------+
|  Review Cards    |<----|  Practice Gen    |<----|  Explanation     |
|  (Anki/Quizlet)  |     |  (problems +     |     |  Engine          |
|                  |     |   solutions)     |     |  (multi-layer)   |
+------------------+     +------------------+     +------------------+

Workflow Examples

Scenario 1: Career Switcher Learning SQL

Input: "I'm a marketing manager learning SQL for data analytics. 
I understand spreadsheets well but never coded."

Output:
## SQL Basics — Spreadsheet-to-SQL Translation

| Spreadsheet Action | SQL Equivalent |
|-------------------|----------------|
| Filter rows       | WHERE clause   |
| Sort column       | ORDER BY       |
| VLOOKUP           | JOIN           |
| Pivot table       | GROUP BY       |

**Practice:** "Write a query to find all customers who spent 
over $100 last month" [solution with spreadsheet parallel]

Scenario 2: Physics Student Struggling with Quantum Mechanics

Input: "I get classical mechanics but quantum superposition 
makes no sense to me"

Output:
## Superposition — Bridge from Classical

**What you already know:** A coin is heads OR tails.
**The quantum twist:** Before you look, the coin is in a 
mathematical state that contains BOTH possibilities with 
probabilities. Not because we don't know — because it 
genuinely hasn't "decided" yet.

[Builds from coin analogy to Dirac notation step-by-step]

Scenario 3: Expert Needing Niche Subtopic

Input: "I'm a senior ML engineer. Explain the lottery ticket 
hypothesis and its implications for model pruning."

Output:
[Expert-level treatment with paper citations, mathematical 
formulation, recent experimental results, and open questions 
in the field. No analogies unless requested.]

Best Practices

• State your current knowledge honestly. The agent calibrates entirely based on your self-assessment and responses. Overstating your level leads to gaps; understating wastes time. Say exactly what you know and what confuses you.

• Ask follow-up questions aggressively. The best learning happens in dialogue. If an analogy does not click, say so. If you want the mathematical formulation after the intuitive explanation, ask for it. The agent adapts in real time.

• Complete practice problems before checking solutions. The agent provides step-by-step solutions, but attempting problems yourself first activates recall pathways. Productive struggle is where real learning happens.

• Export to spaced repetition regularly. After each session, generate Anki or Quizlet cards. Concepts explained today vanish without periodic review. The agent timestamps cards for optimal review scheduling.

• Revisit topics at higher levels over time. Return to previously learned concepts and request expert-level treatment. This spiral approach builds genuine depth rather than superficial familiarity with the material.

Common Issues

• Explanations feel too simple or too complex. Explicitly state your level at the start of each session and correct the agent mid-explanation. Say "go deeper" or "back up a step" as needed. The agent recalibrates immediately.

• Practice problems do not match your curriculum. Specify your course textbook, syllabus, or exam format. The agent can tailor problems to match specific academic standards, AP/IB formats, or professional certifications.

• Anki exports have formatting issues. Ensure your Anki installation supports markdown rendering via an addon. Use the cloze card type for best results. If LaTeX equations break, switch the output_format to plain-text-math.

Privacy & Data Handling

• Local Processing: All explanations are generated in-session. No learner data is stored on external servers. Your knowledge level assessments remain in your local conversation context.
• Data Retention: Session history is retained only in your local Claude Code environment. No learning profiles are transmitted or persisted to any external service.
• Export Options: Full session transcripts can be exported as Markdown, PDF, or structured JSON. Flashcard exports are saved locally before any optional upload to Anki or Quizlet.
• No Tracking: The agent does not track learning analytics externally. All progress data stays in your configured local storage such as Notion, Obsidian, or your local filesystem.