PADH AI is an AI-powered course generation system built around three cooperative LLM agents. Given a topic or raw learning material, the system:

Architects a structured, pedagogically-grounded course blueprint.

Generates rich, format-specific content for every learning event with real-world grounding.

Simulates a struggling student attempting the course and rewrites any confusing sections — automatically.

Most AI course generators produce generic content. PADH AI is different: it uses a self-improving feedback loop where a simulated "weakest student" actively attempts the generated exercises. Any content that the simulated student fails to understand is automatically flagged, its concept gaps are identified, and the Content Agent rewrites it — all in a single pipeline run.

This means every course produced by PADH AI has been stress-tested against a struggling learner before it ever reaches a real student.

Location: architect_agent/

The Architect Agent is the course designer. It takes a raw topic or description and produces a complete course blueprint structured around Gagné's Nine Events of Instruction — a research-backed pedagogical framework used in professional instructional design.

Step 1 — Prompt Refinement & Classification The agent runs the input through a lightweight LLM acting as a "CS Instructional Gatekeeper". This step:

Rejects non-CS topics (cooking, history, etc.) with a friendly message.

Classifies valid CS topics into CODE, MATH, or CONCEPT.

Step 2 — Web Grounding (via Tavily) The agent optionally queries the Tavily Search API to retrieve real-world, up-to-date documentation and examples. This grounds the blueprint in actual current practice.

Step 3 — Blueprint Generation The final LLM call produces a strict JSON blueprint with 9 events. The agent specifically configures Event 3 (Quiz) to focus on prerequisite fundamentals needed for the topic, ensuring a proper knowledge baseline.

Location: content_agent/

The Content Agent takes the blueprint and generates rich, format-specific educational content for every event. It runs all 9 events in parallel for fast generation.

Interactive Quizzes: Generates MCQs with a specific syntax that is parsed into an interactive UI block where students can click options and receive immediate feedback.

Markdown Rendering: All technical content is rendered with full markdown support, including code blocks, tables, and lists.

Tavily Enrichment: Key formats (Lecture, Worked Example, Assessment) are enriched with live search context for maximum accuracy.

Format What it generates hook Vivid real-world scenario + provocative question. objectives_list 5–7 Bloom's taxonomy measurable outcomes. quiz 5 interactive MCQ diagnostic questions testing prerequisites. lecture_with_formula Structured lecture with concept overview and technical frameworks. worked_example Step-by-step case study with scenario and requirements. practice_problem Hands-on scenario with specific tasks and deliverables. feedback_rubric Ideal solution walkthrough + grading rubric table. assessment_task Complex multi-stakeholder case study for mastery check. reflection_essay Personal application prompts and reflection questions.

Location: student_agent/

The Student Agent models the "weakest student" — modelled as a learner who is easily confused by jargon and needs concrete analogies.

Attempt: The student persona attempts to explain the concept in their own words.

Score: An Evaluator LLM scores the comprehension (0.0 to 1.0) and identifies specific concept gaps.

Refine: If the score is below the threshold, the Refiner LLM rewrites the original content to address those specific gaps (adding analogies, simplifying language, etc.).

Blueprint Phase: User input is processed into a 9-event pedagogical roadmap.

Drafting Phase: Content for all events is generated in parallel.

Evaluation Phase: The Student Agent loops through the content to find confusing sections.

Refinement Phase: Confusing sections are automatically rewritten and added to the final course.

UI Rendering: The entire journey is displayed in a premium vertical timeline.

Layer Technology Backend FastAPI LLM Provider Groq (Llama 3.1 & 3.3) Web Search Tavily API Rendering Marked.js (Markdown) Export html2pdf.js (PDF Generation) Styling Vanilla CSS (Premium Dark/Light Beige)

Configure Environment: Create a .env file with your GROQ_API_KEY and TAVILY_API_KEY.

Then open: http://127.0.0.1:8000

Interactive Timeline: A beautiful vertical journey showing your progression through the course blueprint.

Premium Content Cards: Collapsible, markdown-ready cards with format-specific icons and duration tracking.

Live Quiz Block: Interactive MCQs that allow you to test yourself with immediate feedback.

Student Agent Logs: A "Behind the Scenes" section to see the AI's internal evaluation of the course quality.

PDF Export: A one-click button to download the entire course curriculum as a professionally formatted PDF.

PADH AI is built on Gagné's Nine Events of Instruction, ensuring every course follows a proven psychological flow:

Gain Attention

Inform Objectives

Stimulate Recall (Prerequisites)

Present Content

Provide Guidance

Elicit Performance

Provide Feedback

Assess Performance

Enhance Retention