Track
Reasoning Agents (Azure AI Foundry)
Project Name
Remote Interview Fraud Dectector
GitHub Username
Likhith2007
Repository URL
https://github.com/Likhith2007/interview.git
Project Description
What does your project do? The Interview Fraud Detection Agent is an intelligent monitoring solution built on the Microsoft Agent Framework that detects and analyzes suspicious activities during technical interviews. It monitors overlay windows on a candidate's screen and uses a two-stage pipeline to identify unauthorized tools like ParakeetAI, ChatGPT, or remote access software. It runs deterministic pattern matching against 500+ threat signatures, followed by deep contextual reasoning using Azure AI Inference (GPT-4o-mini via GitHub Models).
What problem does it solve? Technical interviews are increasingly vulnerable to AI-assisted cheating and unauthorized collaboration. While traditional proctoring systems can see that overlay windows exist, they lack the intelligence to understand why a specific process is suspicious or what action to take. This leads to false positives or completely missed sophisticated cheats. Our agent solves this by applying real LLM reasoning to evaluate context and severity in real-time.
What are the key features?
LLM-Powered Reasoning: GPT-4o-mini evaluates contextual factors (frequency, obfuscation, spoofing) to generate nuanced fraud explanations, moving beyond simple keyword matching.
Instant Automated Alerting: Automatically generates and sends professional, AI-written email alerts to interviewers the moment a medium or high-risk threat is detected.
Comprehensive Threat Categorization: Built-in recognition for 6 distinct threat categories: AI Interview Tools, Proctoring Cheats, Screen Recorders, Remote Access, DevTools, and VPNs.
Low Latency: Processes local telemetry and generates AI-reasoned alerts in under a second.
Awesome job getting the code pushed to GitHub (https://github.com/Likhith2007/interview.git)!
Demo Video or Screenshots
Demo -- https://youtu.be/mIxYSy14UZ8
Primary Programming Language
Python
Key Technologies Used
Microsoft Agent Framework
Azure AI Inference (via GitHub Models)
Azure OpenAI Service SDK (Python)
GPT-4o-mini (Underlying LLM)
C# / .NET 6.0 (WPF Desktop Client)
Python 3.11 (Agent Backend)
Uvicorn / FastAPI (ASGI Server)
Pydantic (Data Validation & Modeling)
Submission Type
Individual
Team Members
No response
Submission Requirements
Quick Setup Summary
- Set up the Reasoning Agent (Python Backend):
Navigate to the interview-fraud-agent directory.
Run pip install -r requirements.txt.
Create a .env file containing your GITHUB_TOKEN (used for Azure AI Inference via GitHub Models).
Start the agent by running python standalone_server.py. It will listen on http://localhost:8088.
2. Run the Desktop Monitor (C# Frontend):
In a new terminal, navigate to the root directory containing the C# WPF project.
Run dotnet run to launch the Desktop application.
Click the ⚙️ Settings button to configure your sender and recipient email addresses.
Click Start Monitoring.
The C# application will instantly begin scanning for overlay windows. When an overlay is detected, it sends the payload to the Python agent, which uses GPT-4o-mini to analyze the context. If the AI determines the activity is high-risk (e.g., an AI interview assistant or proctoring cheat), it automatically generates and sends an email alert to the interviewer.
Technical Highlights
Agentic Reasoning via Azure AI Inference: Uses GPT-4o-mini (accessed via GitHub Models and the OpenAI SDK) to evaluate behavioral indicators, detection frequencies, and process anomalies to generate intelligent, context-aware fraud assessments.
Two-Stage Analysis Pipeline: Combines lightning-fast deterministic pattern matching (evaluating 500+ known threat signatures) with deep LLM reasoning. This ensures both ultra-low latency and high contextual accuracy.
Microsoft Agent Framework Integration: Built in alignment with the official framework standards, exposing a robust /responses conversational protocol that parses structured JSON payloads.
Cross-Language Interoperability: Seamlessly bridges a C#/.NET 6.0 desktop frontend (which hooks into low-level Windows APIs to scan for top-level overlay windows) with a Python-based FastAPI agent backend.
Automated Action Execution: The agent doesn't just analyze data—it acts. It dynamically formats its AI-generated explanations into professional email alerts and triggers the SMTP pipeline immediately upon detecting high-risk behavior.
Graceful Fallbacks: If the LLM service experiences downtime or network issues, the system automatically falls back to its deterministic rule-based threat classification, ensuring the interview monitoring process is never interrupted.
Challenges & Learnings
No response
Contact Information
email -- ilikhith2007@gmail.com
Country/Region
India
Track
Reasoning Agents (Azure AI Foundry)
Project Name
Remote Interview Fraud Dectector
GitHub Username
Likhith2007
Repository URL
https://github.com/Likhith2007/interview.git
Project Description
What does your project do? The Interview Fraud Detection Agent is an intelligent monitoring solution built on the Microsoft Agent Framework that detects and analyzes suspicious activities during technical interviews. It monitors overlay windows on a candidate's screen and uses a two-stage pipeline to identify unauthorized tools like ParakeetAI, ChatGPT, or remote access software. It runs deterministic pattern matching against 500+ threat signatures, followed by deep contextual reasoning using Azure AI Inference (GPT-4o-mini via GitHub Models).
What problem does it solve? Technical interviews are increasingly vulnerable to AI-assisted cheating and unauthorized collaboration. While traditional proctoring systems can see that overlay windows exist, they lack the intelligence to understand why a specific process is suspicious or what action to take. This leads to false positives or completely missed sophisticated cheats. Our agent solves this by applying real LLM reasoning to evaluate context and severity in real-time.
What are the key features?
LLM-Powered Reasoning: GPT-4o-mini evaluates contextual factors (frequency, obfuscation, spoofing) to generate nuanced fraud explanations, moving beyond simple keyword matching.
Instant Automated Alerting: Automatically generates and sends professional, AI-written email alerts to interviewers the moment a medium or high-risk threat is detected.
Comprehensive Threat Categorization: Built-in recognition for 6 distinct threat categories: AI Interview Tools, Proctoring Cheats, Screen Recorders, Remote Access, DevTools, and VPNs.
Low Latency: Processes local telemetry and generates AI-reasoned alerts in under a second.
Awesome job getting the code pushed to GitHub (https://github.com/Likhith2007/interview.git)!
Demo Video or Screenshots
Demo -- https://youtu.be/mIxYSy14UZ8
Primary Programming Language
Python
Key Technologies Used
Microsoft Agent Framework
Azure AI Inference (via GitHub Models)
Azure OpenAI Service SDK (Python)
GPT-4o-mini (Underlying LLM)
C# / .NET 6.0 (WPF Desktop Client)
Python 3.11 (Agent Backend)
Uvicorn / FastAPI (ASGI Server)
Pydantic (Data Validation & Modeling)
Submission Type
Individual
Team Members
No response
Submission Requirements
Quick Setup Summary
Navigate to the interview-fraud-agent directory.
Run pip install -r requirements.txt.
Create a .env file containing your GITHUB_TOKEN (used for Azure AI Inference via GitHub Models).
Start the agent by running python standalone_server.py. It will listen on http://localhost:8088.
2. Run the Desktop Monitor (C# Frontend):
In a new terminal, navigate to the root directory containing the C# WPF project.
Run dotnet run to launch the Desktop application.
Click the ⚙️ Settings button to configure your sender and recipient email addresses.
Click Start Monitoring.
The C# application will instantly begin scanning for overlay windows. When an overlay is detected, it sends the payload to the Python agent, which uses GPT-4o-mini to analyze the context. If the AI determines the activity is high-risk (e.g., an AI interview assistant or proctoring cheat), it automatically generates and sends an email alert to the interviewer.
Technical Highlights
Agentic Reasoning via Azure AI Inference: Uses GPT-4o-mini (accessed via GitHub Models and the OpenAI SDK) to evaluate behavioral indicators, detection frequencies, and process anomalies to generate intelligent, context-aware fraud assessments.
Two-Stage Analysis Pipeline: Combines lightning-fast deterministic pattern matching (evaluating 500+ known threat signatures) with deep LLM reasoning. This ensures both ultra-low latency and high contextual accuracy.
Microsoft Agent Framework Integration: Built in alignment with the official framework standards, exposing a robust /responses conversational protocol that parses structured JSON payloads.
Cross-Language Interoperability: Seamlessly bridges a C#/.NET 6.0 desktop frontend (which hooks into low-level Windows APIs to scan for top-level overlay windows) with a Python-based FastAPI agent backend.
Automated Action Execution: The agent doesn't just analyze data—it acts. It dynamically formats its AI-generated explanations into professional email alerts and triggers the SMTP pipeline immediately upon detecting high-risk behavior.
Graceful Fallbacks: If the LLM service experiences downtime or network issues, the system automatically falls back to its deterministic rule-based threat classification, ensuring the interview monitoring process is never interrupted.
Challenges & Learnings
No response
Contact Information
email -- ilikhith2007@gmail.com
Country/Region
India