ContentKosh AI Content Agent

LangGraph + OpenAI | Stateful Multi-Step Educational Content Generation

Built as a proof-of-work demo for the AI Agent Developer role at ContentKosh.

---

What It Does

A stateful, multi-step AI agent that automates educational content creation for coaching institutes. Given a topic and exam target, it executes a 4-node LangGraph pipeline:

[Topic Input]
     ↓
[Node 1] generate_notes    → Exam-ready study notes (200–250 words)
     ↓
[Node 2] generate_mcqs     → MCQ test set with answers + explanations
     ↓
[Node 3] review_quality    → Factual accuracy + MCQ quality check (JSON)
     ↓
[Node 4] assemble_package  → White-label-ready final content package
     ↓
[Output Package]

Supported exams: UPSC · SSC CGL · CUET · Class 10 Boards · State PCS · NEET · JEE

---

Why This Architecture

Feature	Simple LLM Call	This Agent
State shared across steps	❌	✅ MCQs use notes as context
Quality gate before output	❌	✅ Node 3 flags errors before delivery
Modular — swap any node	❌	✅ Add translation, difficulty, plagiarism nodes
Scales to parallel branches	❌	✅ LangGraph supports async + parallel nodes
Cost control	❌	✅ Per-node token caps + session rate limiting

---

Production Features

• Structured logging (logging module) at every pipeline node
• Input validation with descriptive ValueError / EnvironmentError messages
• Per-node max_tokens caps — prevents runaway OpenAI spend
• Session-level rate limiting in the Streamlit UI (10 req/session default)
• Graceful JSON fallback — Node 4 handles malformed quality review responses
• Non-root Docker user — security-hardened container
• GitHub Actions CI — runs tests + basic secret scan on every push
• MIT licensed

---

Quick Start

Option 1: Local (Python)

git clone https://github.com/SKcoder6344/contentkosh-agent
cd contentkosh-agent

pip install -r requirements.txt

export OPENAI_API_KEY="your-key-here"

# Streamlit UI
make run

# CLI demo
make demo

Option 2: Docker

docker build -t contentkosh-agent .
docker run -p 8501:8501 -e OPENAI_API_KEY="your-key-here" contentkosh-agent
# → Open http://localhost:8501

---

Folder Structure

contentkosh-agent/
├── agent.py                  # Core pipeline (4 LangGraph nodes)
├── app.py                    # Streamlit UI
├── requirements.txt          # Pinned dependencies
├── Makefile                  # Dev shortcuts (run, test, docker-build…)
├── Dockerfile                # Multi-stage production container
├── LICENSE                   # MIT
├── metrics.csv               # Manual vs. agent performance comparison
├── tests/
│   └── test_agent.py         # pytest unit tests (mocked LLM)
└── .github/
    └── workflows/
        └── ci.yml            # GitHub Actions: test + secret scan

---

API Usage

from agent import run_agent

output = run_agent(
    topic="Photosynthesis",
    subject="Biology",
    exam_target="NEET",
    num_mcqs=5
)
print(output)

---

Performance

Metric	Manual Process	This Agent
Content package creation	3–4 hours	~45 seconds
MCQ generation per topic	15–20 min	~5 seconds
Factual accuracy check	Manual review	Automated (Node 3)
Scale (topics/day)	10–15	100+

---

Roadmap Alignment

Current Node	Future Extension	Business Value
generate_notes	generate_flashcards	Spaced repetition feature
review_quality	plagiarism_check	Originality compliance
assemble_package	translate_to_hindi	Regional language expansion
New: difficulty_adjuster	Scale JEE vs Class 10	Single agent, multiple exams

---

Running Tests

make test
# → pytest tests/ -v --cov=agent

---

Author

Sujal Kumar Nayak
GitHub: github.com/SKcoder6344
Email: nayaksujalkumar@gmail.com