Skip to content

Founder · Researcher · Student

Faris Allafi

I build efficient machine-learning systems for hardware that has no business running them. At DimbaLabs I’m designing DIMBA-II, a Mamba-2 state space model with latent diffusion aimed at on-device reasoning under extreme quantization. I work close to the metal: Rust, embedded targets, and a lot of benchmarking.

Focus
On-device reasoning
Target
Sub-2-bit MoE
Based in
Abu Dhabi, UAE

Experience

DimbaLabs / DIMBA-II

2025 – Present

Founder, Researcher & Student

  • Designing DIMBA-II: a Mamba-2 state space model with latent diffusion for on-device reasoning, targeting extreme quantization and sub-2-bit MoE configurations.

  • Running benchmarking and training experiments on Kaggle GPU infrastructure; evaluating inference latency against semantic coherence tradeoffs.

  • In active discussions for compute and capital partners.

Research

Speaking & Competitions

  • Speaker

    Make it in the Emirates

    Speaker at the UAE industrial innovation event.

  • Qualifier

    World Scholar's Cup

    Qualified for the Global Round.

  • Winner

    Khalifa University Cybersecurity Competition

    First place overall.

  • Lead

    FIRST Tech Challenge Robotics

    Team leader (FTC).

Education

West Yas Academy (Aldar Education)

Abu Dhabi, UAE

Grade 7Academic Scholarship

Technical Skills

Languages
RustPythonGoTypeScriptC
AI / ML
Mamba SSMsLatent diffusionModel quantizationExtreme compressionBenchmarking
Embedded
ESP32-S3Sensor calibrationData acquisition pipelines
Systems
AxumTokioAsync RustAgent architectureCron scheduling
Tools
GitKaggleLinux (Arch)tmux
Other
Speed reading (~5000 WPM)Technical writing

Contact

Email

Elsewhere

GitHubdevnull37X@farisallafiLinkedInfaris-allafi