README.md

# Interviewer

A multi-agent AI powered coding interviewer built using AutoGen.

The system:
- generates coding problems
- writes them into files
- lets candidates solve them locally
- evaluates the solution
- asks follow-up questions
- scores the candidate
- stores results in Excel

Basically your own AI technical interviewer.

---

# Features

- Multi-agent interview orchestration
- Real coding interview flow
- Topic-based DSA question generation
- MCP filesystem integration
- Code evaluation + scoring
- Excel result storage
- Prompt injection resistance
- Cross-platform support (Windows/Mac/Linux)

---

# Tech Stack

- Python 3.13+
- AutoGen
- MCP Filesystem Server
- OpenRouter / OpenAI / Groq compatible
- Rich CLI
- Questionary
- OpenPyXL

---

# Installation

## 1. Clone the repo

```bash
git clone <your-repo-url>
cd interviewer

---

2. Install uv

If you don't have uv installed...

Well, I got you covered, my friend

Windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Mac/Linux

curl -LsSf https://astral.sh/uv/install.sh | sh

Docs: https://docs.astral.sh/uv/

---

3. Sync dependencies

uv sync

---

4. Create .env

Copy everything from:

.env.example

into:

.env

---

API Keys

OpenRouter (Recommended)

Go here:

https://openrouter.ai/

Create your own API key.

It's free up to a reasonable limit.

And secondly...

why would I give you mine? hehehe

---

Running the Project

uv run python -m main

---

Supported Models

You can swap models easily inside interviewer.py.

---

Free Models

DeepSeek Free

client = OpenAIChatCompletionClient(
    model="deepseek/deepseek-chat:free",
    base_url="https://openrouter.ai/api/v1",
    api_key=os.getenv("OPENROUTER_API_KEY"),
    model_info={
        "vision": False,
        "function_calling": True,
        "json_output": True,
        "structured_output": True,
        "family": "unknown",
    },
)

---

Llama 3.3 70B (Groq)

Very fast.

client = OpenAIChatCompletionClient(
    model="llama-3.3-70b-versatile",
    base_url="https://api.groq.com/openai/v1",
    api_key=os.getenv("GROQ_API_KEY"),
    model_info={
        "vision": False,
        "function_calling": True,
        "json_output": True,
        "structured_output": True,
        "family": "llama",
    },
)

---

Gemini Flash

Cheap and good.

client = OpenAIChatCompletionClient(
    model="google/gemini-2.0-flash-001",
    base_url="https://openrouter.ai/api/v1",
    api_key=os.getenv("OPENROUTER_API_KEY"),
    model_info={
        "vision": False,
        "function_calling": True,
        "json_output": True,
        "structured_output": True,
        "family": "gemini",
    },
)

---

Paid Models

GPT-4o

Best overall quality.

client = OpenAIChatCompletionClient(
    model="gpt-4o",
    api_key=os.getenv("OPENAI_API_KEY"),
)

---

GPT-4.1 Mini

Cheap + strong reasoning.

client = OpenAIChatCompletionClient(
    model="gpt-4.1-mini",
    api_key=os.getenv("OPENAI_API_KEY"),
)

---

Claude Sonnet

Excellent interviewer behavior.

client = OpenAIChatCompletionClient(
    model="anthropic/claude-3.7-sonnet",
    base_url="https://openrouter.ai/api/v1",
    api_key=os.getenv("OPENROUTER_API_KEY"),
    model_info={
        "vision": False,
        "function_calling": True,
        "json_output": True,
        "structured_output": True,
        "family": "claude",
    },
)

---

Project Structure

interviewer/
│
├── main.py
├── interviewer.py
├── problems/
├── interview_results.xlsx
├── .env.example
├── pyproject.toml
└── README.md

---

How It Works

1. Candidate selects:

   • difficulty
   • DSA topics

2. AI interviewer:

   • generates a coding problem
   • writes it into /problems

3. Candidate solves problem locally

4. Evaluator agent reviews the code

5. Scorer agent:

   • assigns score
   • stores results in Excel

---

Security Features

• Prompt injection resistance
• Topic-constrained generation
• Filesystem sandboxing
• Agent role enforcement
• Controlled evaluation flow

---

Future Improvements

• Web UI
• Voice interviews
• Live collaborative coding
• Resume-based question generation
• Multi-round interviews
• System design rounds
• Docker support

---

Disclaimer

If the AI rejects your solution brutally...

that's just realistic interview prep :)

Additional Information/Advice

• I have noticed that using a paid model gives me a better result than free ones, as hallucinations are less
• I recommend using OpenAI or Google Gemini for this
• Avoid using laptops that you are not an admin of, as they may block calls to open-source LLM models