Skip to content
View cataluna84's full-sized avatar
:octocat:
Founder & CEO
:octocat:
Founder & CEO
50 followers · 329 following

Achievements

Achievement: Pull Sharkx2Achievement: Pair Extraordinairex2Achievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pull Sharkx2Achievement: Pair Extraordinairex2Achievement: YOLOAchievement: Quickdraw

Highlights

Organizations

@Cohere-Labs-Community

Block or report cataluna84

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please donโ€™t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this userโ€™s behavior. Learn more about reporting abuse.

Report abuse
cataluna84/README.md

Hi, I'm Mayank Bhaskar ๐Ÿ‘‹

Founder & CEO โ€” OM Enterprises ยท AI Research & Engineering Labs Senior ML Engineer ยท AI Researcher ยท 14+ years building production AI systems

๐ŸŒ Portfolio: cataluna84.github.io โ€” full project narrative, ARC-AGI work, Kaggle competitions, experience timeline ๐Ÿ“‚ Portfolio source: cataluna84/cataluna84.github.io

Portfolio LinkedIn Kaggle HuggingFace X

๐Ÿ”ญ What I'm doing now

Building ubiquitous, interactive intelligence that runs on multimodal data (audio, text, image, video) with high throughput and low latency.

My research spans Computer Vision, State Space Models (S4 / S5 / Mamba), Multimodal Models, Reinforcement Learning, Mechanistic Interpretability, and LLM compression โ€” with applied experience deploying on-device (iOS), on edge (DeepStream SDK), and at scale (CUDA/Triton).

๐Ÿ† Highlights
๐Ÿฅˆ ARC Prize 2024 24th / 1,427 teams โ€” Silver Medal
๐Ÿฅ‰ ARC Prize 2025 87th / 1,455 teams โ€” solo Bronze Medal
๐Ÿ”„ ARC Prize 2026 Active competitor
๐Ÿฅ‰ HF Flax/JAX Sprint 3rd place โ€” CLIP-RSICD (satellite images)
๐Ÿ“Š Kaggle Competitions Expert ยท rank ~1,532 / 202,876 ยท 1 Silver + 3 Bronze across 21+ competitions
๐Ÿค— HuggingFace 14 models published (NER, text generation, image generation)

๐Ÿš€ Featured projects

๐Ÿ”ฌ LID โ€” Layer-Wise Multilingual Language ID โ€” Layer-wise dynamics of multilingual language identification across 67 languages in compact foundation models. PyTorch ยท LoRA ยท Macro F1 0.97+

๐Ÿง  ARC Prize 2026 (ARC-AGI-3) โ€” Public lab notebook for the $850K ARC-AGI-3 competition. Agent zoo, FORGE, BFS, CNN, frame segmenter.

๐ŸŽต TinyAya: Simultaneous Translation โ€” A set of 6 public GitHub repos having benchmarks for fine-tuning neural speech codecs (Mimi, DualCodec, Kanade) on low-resource languages with 8 optimizers + W&B Bayesian sweeps. Plus, the complete end to end model weights trained on v6e-16 TPUs

๐Ÿ›ก๏ธ Crosslingual Emergent Misalignment โ€” Cohere Labs research on how safety guardrails degrade across high- and low-resource languages. Published as a virtual poster at the ICML 2026 Mechanistic Interpretability Workshop.

๐Ÿš˜ Tiny Aya Under the hood โ€” This project investigates how Tiny Aya processes information across languages by analyzing how representations evolve across model layers.

๐ŸŽจ Generative Deep Learning โ€” VAE, GAN, and Diffusion implementations for image synthesis.

๐Ÿ Jamba โ€” Hybrid Transformerโ€“Mamba โ€” PyTorch implementation exploring SSM + attention for infinite-context LMs.

๐ŸŒ World Models (Ha & Schmidhuber) โ€” VAE + MDN-RNN + CMA-ES for RL in the OpenAI Car Racing environment.

โ†’ Full project list with metrics, write-ups, and demos: cataluna84.github.io

๐Ÿค Open source contributions

  • Google DeepMind โ€” penzai (JAX neural network library)
  • TransformerLens โ€” TransformerLens (mechanistic interpretability)
  • Kyutai Labs โ€” moshi (open-source speech-text foundation model)

๐Ÿ› ๏ธ Tech stack

Deep Learning & Research PyTorch JAX Mamba/S4/S5 Transformers CUDA Triton

ML Engineering & MLOps W&B HuggingFace Docker Quantization FlashAttention

Computer Vision & Edge DeepStream iOS CLIP

Languages & Tools Python C++ uv Linux Git

๐ŸŒฑ Community & leadership

  • Cohere Labs โ€” Community Lead (Computer Vision sub-field) & Researcher
  • TWiML (This Week in ML) โ€” Contributor Lead, 5+ years moderating weekly GenAI meetups
  • Lucknow AI Labs โ€” Core organizer, Research Paper Team lead
  • Machine Learning Tokyo (MLT) โ€” Contributor (2019โ€“2023)
  • Yannic Kilcher Discord โ€” Daily/weekly paper discussions

Conferences attended (remotely): ICLR ยท ICML ยท CVPR ยท NeurIPS

๐Ÿ“ˆ GitHub stats
Contributions (last year) 566
Commits (last year) 517
Pull requests merged 37
Stars earned 33
Followers 49
Repositories 44 (31 personal, 13 org & collaborative)

Languages โ€” share of the 44 repositories each language appears in:

Python            โ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆ   54.5%  (24 repos)
Jupyter Notebook  โ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘   45.5%  (20 repos)
Shell             โ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘   31.8%  (14 repos)
Java              โ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘   18.2%  (8 repos)
Makefile          โ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘    9.1%  (4 repos)
Swift             โ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘    9.1%  (4 repos)
TeX               โ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘    9.1%  (4 repos)
Dockerfile        โ–ˆโ–ˆโ–ˆโ–ˆโ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘โ–‘    9.1%  (4 repos)

Weighted by repository count, not bytes: notebook files embed their rendered outputs, so byte-weighted shares measure figure size rather than code. Includes organization and collaborative repositories. Last refreshed 2026-11-01.

๐Ÿ“ซ Get in touch

Open to Senior ML Engineer ยท Research Engineer ยท CV Engineer ยท Senior SDE roles, research collaborations, and advisory work.

The portfolio site is the canonical source โ€” this README is a GitHub-native summary that links back to it.

Pinned Loading

  1. tiny-aya-simultaneous-translation/model tiny-aya-simultaneous-translation/model Public

    TinyAya Stage 2: TRโ†”HI speech-to-speech translation at scale

    Python 2 1

  2. arc-agi-3 arc-agi-3 Public

    Public lab notebook for the Kaggle ARC Prize 2026 - ARC-AGI-3 competition: agents, experiments, and tooling. Apache 2.0.

    Jupyter Notebook 4

  3. lid lid Public

    Layer-wise dynamics of multilingual language identification (67 languages) in compact foundation models โ€” training, layer-wise inference, and optimization-strategy benchmarks.

    Jupyter Notebook 1

  4. rsk2327/Tiny-Aya-Under-the-hood rsk2327/Tiny-Aya-Under-the-hood Public

    This project investigates how Tiny Aya processes information across languages by analyzing how representations evolve across model layers.

    Jupyter Notebook 5 1

  5. Cohere-Labs-Community/crosslingual-emergent-misalignment Cohere-Labs-Community/crosslingual-emergent-misalignment Public

    A community research project investigating emergent misalignment in multilingual language models.

    Jupyter Notebook 7 2

  6. Generative_Deep_Learning Generative_Deep_Learning Public

    Generative Modeling

    Jupyter Notebook 4 1

โŒ‚