Spectral AI Lab is a small, independent AI lab exploring generative AI, foundation models and what lies beyond them.

Spectral was born from our everyday conversations about AI. Behind it are Dr. Ichrak Mokhtari and Seif Attoui, two AI enthusiasts with deep roots in the field. No large team, no giant cluster: just two curious minds contributing to AI research with the means we have. We follow the state of the art closely, put new models and architectures to the test, and love taking ideas from first intuition to working model. And because sharing is caring, everything we build is open.

What we believe

The future of AI will not be written by scale alone. Real progress also comes from efficiency, careful data craftsmanship, the right inductive biases, and radical transparency in how results are measured and shared.

What we explore

• Small models, big ideas: compact architectures that punch above their weight and run on everyday hardware
• Data craftsmanship: curated, high-quality data as a lever as powerful as scale
• Inductive biases: architectural choices that help models learn and reason with less
• Taking open models further: continued pretraining, fine-tuning and reinforcement learning
• Transparent evaluation: open weights, reproducible benchmarks, no cherry-picking

Featured

Connect

GitHub · Ichrak · Seif · Questions and ideas are welcome in our model discussions.

_{In mathematics, a spectrum reveals the hidden structure of a system. Open science, shared freely.}