I'm Siradj Mounir Lamri, an AI & Software Engineer pursuing advanced studies in Artificial Intelligence & Applications in Algeria 🇩🇿.
I specialize in small, fast inference models (System-1 AI), high-performance ML backends, and full-stack AI engineering.

Active contributor to high-impact, fast-growing AI & ML repositories.

An automated synthesis & rapid adaptation engine that turns cold-start task criteria into specialized, calibrated sub-35ms Laya decision models in 30 seconds.

• 🚀 Automated Synthetic Data: Generates diverse multi-task datasets and adversarial hard-negatives to immunize small models against keyword & negation traps.
• ⚡ Rapid Head Adaptation: Freezes encoder backbones and trains decision heads in ~15–25 seconds with post-hoc temperature calibration (driving ECE < 0.08).
• 🎨 Interactive UI: Built-in Gradio interface for 1-click bootstrapping and real-time sub-35ms live inference playgrounds.
• ☁️ Live on HuggingFace Spaces with ZeroGPU (A10G) — try it at Lamri26/laya-ignite.

Laya is a non-autoregressive AI engine that outputs strict JSON probabilities with sub-35ms latency, eliminating LLM token latency for deterministic workflows.

Key technical details — PR #387 (Batch Inference Endpoint):

• Designed the BatchDecideRequest / BatchDecisionResponse Pydantic schema for N-state fan-out over a shared question bank.
• Implemented POST /v1/systemone/batch in FastAPI with single-forward-pass batching to maximize GPU utilization.
• Added concurrency admission gating (503 Service Unavailable) to protect the server under load.
• Enforced strict option limits (max 20 options per question) with descriptive validation errors.
• Full test suite: 62 passing tests including new batch endpoint cases, rebased cleanly on v0.3.21 (9d95567).

Other contributions:

• 🇩🇿 Sole Algerian Contributor to the core Laya ecosystem.
• 📦 TypeScript/JavaScript SDK (laya-sdk): Drop-in replacement for @typesafe-ai/sdk with full type safety and decision builders (choice, score, noul).
• 🎨 Brand Identity: Conceptualized the "converging dots" motif adopted as the geometric basis for Laya's official branding.

Python	TypeScript	JavaScript	PyTorch	FastAPI	Angular
Docker	Linux	PostgreSQL	MySQL	Git	GitHub

 

• 🤖 System-1 AI & Small Specialized Models: Low-latency classification, routing, and calibrated decision heads.
• ⚡ High-Throughput ML Serving: Batching strategies, concurrent admission gates, and GPU kernel optimizations.
• 🌐 Full-Stack AI Products: Bridging deep learning research with production-grade web APIs and client SDKs.
• 🎯 Reliable AI Guardrails: Structured outputs, schema validation, and defensive API limits.

"Build things. Learn deeply. Make them useful."