Ali Janati.

Ali Janati.

Research in deep learning.

Efficient training and inference for large models, real-time speech-to-speech systems, and interpretability: measuring which parts of a model are doing the work, and using that to make models smaller and faster.

I am a senior machine learning scientist at a startup in San Francisco, where I work on world models and speech. I studied at Columbia and Mines Paris.

Portrait of Ali Janati.

Recent

Papers.

Most of my work asks which parts of a large model are actually doing the work. Recent papers span expert pruning in mixture-of-experts models, training dynamics after grokking, uncertainty in multimodal recognition, and machine learning for epigenomics.

Earlier research, with Prof. Pierre Gentine’s group at Columbia: fine-tuned the IBM-NASA geospatial vision transformer to segment wind-damaged forest in satellite imagery.

Articles.

Write-ups of the architectures I reimplement from scratch in PyTorch-Playground: what is new in each model, how I implemented it, and the code that matters.

Open source.

  • PyTorch-Playground

    From-scratch PyTorch reimplementations of modern architectures: Llama 3.1 8B, GPT-2, gpt-oss-20B, Qwen3-80B-A3B, Gemma 270M, and Magistral Small, plus speculative decoding and quantization-aware training.

  • Medical-Whisper-Large-v3

    Whisper large-v3 fine-tuned for medical conversation: 14% lower word-error rate on validation, released as open weights. 35,000+ downloads on Hugging Face.

  • MoE pruning models

    Open artifacts behind the router-sensitivity paper: Mixtral-8x7B and Qwen1.5-MoE variants with experts pruned and ablated, along with the SFT datasets used to recover performance. The Mixtral series is curated as a collection; the full set lives on the profile.

  • muon-grokking

    The experiment suite behind the post-grokking-collapse paper: Muon and AdamW sweeps on modular arithmetic, freezing and unembedding-rate interventions at the representation-readout interface, and a generality suite across moduli, widths, and training configurations.

  • GAN_Monet_Style

    A CycleGAN that transfers Claude Monet’s painting style to photographs.