Skip to main content
ADAmit Divekar
Back to Blogs
AI
Multi-Agent Systems
Gemini
Python
EdTech

Reverse-Engineering Exams: A Deep Dive into Professor Profiler

Amit Divekar

Reverse-Engineering Exams: A Deep Dive into Professor Profiler

In the world of academia, the difference between a good grade and a great one often isn't just about how much you study—it's about studying the right things. Enter Professor Profiler, a Hierarchical Multi-Agent System (HMAS) designed to mimic the cognitive process of an expert tutor.

By orchestrating specialized AI agents powered by Google Gemini 2.5, this tool doesn't just read exam papers; it reverse-engineers them to decode the psychology of the professor and formulate actionable study plans.

The Hub-and-Spoke Architecture

Building a system that can "think" like a tutor requires more than a single prompt. We utilized a Hub-and-Spoke architecture, creating a directed acyclic graph (DAG) of agent execution managed by a central orchestrator.

At the center sits the Root Agent, acting as the project manager. It delegates tasks to specialized sub-agents, ensuring that each part of the analysis is handled by an expert model suited for that specific cognitive load.

100%
flowchart TD
    subgraph Orchestration_Layer ["🧠 Orchestration Layer"]
        Runner[<b>Runner</b><br/><i>State Management</i>]
        Memory[(<b>Memory Bank</b><br/><i>JSON Persistence</i>)]
    end

    subgraph Agent_Layer ["🤖 Agent Hierarchy"]
        Root[<b>ROOT AGENT</b><br/><i>Gemini 2.5 Pro</i><br/>The Project Manager]
        
        subgraph Workers ["Specialized Sub-Agents"]
            Taxonomist[<b>Taxonomist</b><br/><i>Gemini Flash</i><br/>Classification]
            Trend[<b>Trend Spotter</b><br/><i>Gemini Pro</i><br/>Analysis]
            Strat[<b>Strategist</b><br/><i>Gemini Thinking</i><br/>Planning]
        end
    end

    Runner --> Root
    Root --Delegates--> Taxonomist
    Root --Delegates--> Trend
    Root --Delegates--> Strat

Meet the Agents

The system's power lies in its distinct agent personas, each utilizing a version of Gemini optimized for their specific role:

  1. The Taxonomist (Classifier) Powered by gemini-2.0-flash-exp for speed, this agent acts as the meticulous grader. It reads every question and tags it with a topic and its Bloom's Taxonomy level (e.g., Analyze, Evaluate).

  2. The Trend Spotter (Analyst) Using gemini-2.0-pro-exp for its large context window, this agent acts as the data scientist. It identifies frequency distributions and "curveball" questions that deviate from historical norms.

  3. The Strategist (Coach) Leveraging the reasoning capabilities of gemini-2.0-flash-thinking-exp, this agent synthesizes the data to output the final Safe Zones (high reward, low effort) and Drop Lists.

From Raw PDF to Study Plan

The execution pipeline transforms a raw exam PDF into a comprehensive report through four distinct phases: ingestion, classification, visualization, and strategy formulation.

100%
sequenceDiagram
    autonumber
    actor Student
    participant Root as 🧠 Root Agent
    participant Tax as 🏷️ Taxonomist
    participant Strat as 🎯 Strategist

    Student->>Root: "Analyze Physics_2024.pdf"
    
    note right of Root: Phase 1 & 2: Classification
    Root->>Tax: "Classify these questions"
    Tax-->>Root: JSON List of Classified Questions

    note right of Root: Phase 3: Visualization
    Root->>Root: Generate Charts (Matplotlib)

    note right of Root: Phase 4: Strategy
    Root->>Strat: "Identify Safe Zones"
    Strat-->>Root: Final Recommendations

    Root-->>Student: Report + Images + Plan

Built for Production

We designed Professor Profiler not just as a demo, but as a reference implementation for production-grade agentic systems. Key technical features include:

  • Observability: Integrated structured logging with correlation IDs and distributed tracing to track requests across the agent hierarchy.
  • State Management: An in-memory session service that supports long-running operations, allowing the system to pause and resume analysis.
  • Custom Tooling: A robust tool layer handling PDF text extraction (pypdf) and dynamic visualization generation (matplotlib).

Try It Yourself

The project creates a flexible foundation for any document analysis pipeline. Here is how you can initialize the runner and analyze a file programmatically:

import asyncio
from google.genai import types
from profiler_agent.agent import root_agent
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService

async def main():
    # Initialize memory
    session = InMemorySessionService()
    
    # Initialize runner
    runner = Runner(agent=root_agent, session_service=session)
    
    # Execute
    print("🤖 Agent is thinking...")
    async for event in runner.run_async(
        user_id="prof_user",
        session_id="sess_01",
        new_message=types.Content(
            role="user",
            parts=[types.Part.from_text("Analyze the midterms.")]
        )
    ):
        if event.is_final_response():
            print(f"\n🎓 Final Answer:\n{event.content.parts[0].text}")

if __name__ == "__main__":
    asyncio.run(main())

By combining the reasoning power of Large Language Models with structured cognitive architectures, Professor Profiler demonstrates that AI can be more than just a chatbot—it can be a strategic partner in learning.

Explore the full source code and contribute to the project on GitHub: Professor Profiler.

🚀

Connect With Me

If you found this helpful, let's connect! I share more insights on AI, multi-agent systems, and software development.

  • GitHub: @amitdevx - Check out my projects and code
  • LinkedIn: Amit Divekar - Let's connect professionally

Feel free to star the repos, share your thoughts, or reach out for collaboration!