A modular, LangChain + LangGraph multi-agent system that researches a topic, generates quizzes, and explains answers, with Streamlit UI and LangSmith tracing.

---

Problem Statement

• Many students struggle to find concise explanations, relevant practice questions, and clear reasoning for topics—all in one place.
• Existing tools can be overwhelming or lack interactive, testable learning loops.
• There is a need for a cohesive pipeline from research → quiz → explanation, with transparent orchestration and observability.

---

Solution Overview

• Multi-agent architecture with three specialized agents:
  • Research Agent: Produces clear, concise explanations and practical examples.
  • Quiz Agent: Generates relevant quiz questions with answers.
  • Explainer Agent: Provides step-by-step clarifications for each answer.
• LangGraph orchestration with in-memory checkpointing and thread ID support.
• Streamlit UI for a modern, interactive front-end.
• LangSmith integration for tracing, debugging, and observability.

---

Key Features

• Research → Quiz → Explain pipeline with modular agents.
• Wikipedia and Math tool integration.
• Groq LLM support, model settable via .env.
• Streamlit UI with customizable theme.
• Deterministic fallbacks—never display empty outputs.
• LangSmith tracing and experiment tracking.

---

Architecture

• LLM: Groq (configurable via GROQ_MODEL).
• Orchestration: LangGraph + MemorySaver checkpointing.
• Agent Framework: LangChain.
• UI: Streamlit.
• Tracing & Observability: LangSmith.
• Utilities: Wikipedia API, basic math evaluation.

---

Components

Agents

Research Agent

• Input: topic (+ Wikipedia summary when available)
• Output: concise explanation (80–150 words) + 2 examples
• LLM settings: Groq, temperature 0.2

Quiz Agent

• Input: research summary
• Output: 5 questions with answers in “Q: … A: …” format
• LLM settings: Groq, temperature 0.4

Explainer Agent

• Input: each Q/A line
• Output: step-by-step explanation
• LLM settings: Groq, temperature 0.3

Tools

• Wikipedia Tool: Fetches concise summaries using the Wikipedia REST API.
• Math Tool: Minimal, restricted eval for basic arithmetic.

Orchestrator

• Flow: research → quiz → explain → END
• Checkpointer: in-memory
• API:
  • run_session(topic): Executes the entire pipeline and returns results.
  • orchestrate_session(): CLI wrapper.
• Reliability: Strong prompting and retry/guard logic; deterministic fallbacks to avoid empty outputs.

---

User Interface

• Sidebar: Topic input, model display (auto-set from .env), run button.
• Tabs:
  • Research
  • Quiz
  • Explanations
  • Raw (shows returned JSON)
• Theming: Custom Streamlit palette via configuration.

---

Configuration

Environment Variables (.env)

GROQ_API_KEY=your_groq_api_key_here
GROQ_MODEL=llama-3.3-70b-versatile
LANGSMITH_API_KEY=your_langsmith_key_here
LANGSMITH_PROJECT=edu-sync-agents
LANGCHAIN_TRACING_V2=true
LANGCHAIN_ENDPOINT=https://api.smith.langchain.com

Easily switch Groq models by changing GROQ_MODEL—no code edits needed.

---

Installation & Running

0. Clone Repository

git clone https://github.com/narevignesh/edu-sync-agents.git

### 1. Setup Virtual Environment
python -m venv env

# Activate (Windows)
env\Scripts\Activate

# Activate (Unix/Mac)
source env/bin/activate

pip install -r requirements.txt
cp .env_example .env
# Fill in .env with your keys

2. CLI Run

python main.py

3. Streamlit UI

streamlit run app.py

---

Data Flow

1. User enters topic in UI.
2. Research Agent composes explanation using LLM and Wikipedia summary.
3. Quiz Agent generates 5 Q/A pairs from the research summary.
4. Explainer Agent provides detailed explanations for each Q/A pair.
5. UI displays all sections and raw output; traces are sent to LangSmith.
6. Fallbacks are used if any step fails to ensure continuous output.

---

Reliability & Fallbacks

• LLM prompts are guarded and responses are retried as needed.
• If any model response is empty:
  • Research: Uses Wikipedia summary or templated topic overview.
  • Quiz: Provides deterministic Q/A pairs for the topic.
  • Explain: Uses a generic rationale template.
• If LangGraph returns empty results, a sequential fallback pipeline is invoked.

---

Security and Safety

• All API keys and secrets are loaded only from environment variables in .env (never hard-coded).
• math_tool uses restricted evaluation for security.
• Wikipedia tool only uses public, read-only API requests.

---

Limitations

• Fallbacks are intentionally generic, serving as last-resort outputs.
• Wikipedia coverage may be limited for certain niche topics.
• No persistent data storage, memory checkpointer only.

---

Validation and Testing

• Manual validation across diverse topics.
• Trace review in LangSmith to confirm proper agent handoffs and tool usage.
• Sanity checks for both Wikipedia and math tool behavior.

---

Performance Notes

• Groq LLMs deliver low-latency completion.
• You can switch LLM models for optimal trade-off between quality and speed.
• Add Wikipedia summary caching to boost speed and handle API limits.

---