README.md

---
title: Lex Fridman AI Interviewer
emoji: 🎙️
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.23.0
app_file: app.py
pinned: false
tags:
  - interviewer
  - nemotron
  - lex-fridman
---

# 🎙️ Lex Fridman AI Interviewer

AI interviewer in the style of Lex Fridman, powered by **NVIDIA Nemotron 3 Nano 4B** via HF Inference API.

The model asks deep, thoughtful questions — you are the guest. Paste context about yourself for more relevant interviews.

## Model

- **Base**: [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16)
- **Architecture**: Hybrid Mamba-2 + Attention (38 Mamba layers + 4 Attention layers)
- **Inference**: HF Inference API (ZeroGPU local deployment blocked by mamba-ssm CUDA kernel requirements)

## Known Limitation

ZeroGPU deployment with local model was attempted but the hybrid Mamba-2 architecture requires
`mamba_ssm` CUDA Triton kernels (`rmsnorm_fn`) that can't be compiled in the Space build environment.
A pure PyTorch mock was validated locally but fails on ZeroGPU due to device placement / timeout issues.
See `docs/ZEROGPU_LIMITATION.md` in the project repo for details.

## Project

Part of the [Lex Fridman Interviewer Project](https://huggingface.co/bobber/lex-fridman-interviewer-project).