Overview

Awras is an end-to-end data annotation and dataset management platform designed to accelerate the training of multilingual AI models. It streamlines the complex process of crowdsourcing and refining linguistic data—specifically focusing on translation tasks and dictionary curation—while providing an intuitive, gamified experience for annotators.

Key Features & Achievements

• Advanced Annotation Engine: Developed a specialized backend to handle high-throughput dataset management, allowing admins to distribute translation and dictionary tasks to a workforce of annotators.
• Annotator Gamification: Implemented a real-time leaderboard and contribution tracking system to incentivize data quality and user engagement.
• Enterprise-Grade Identity (IAM): Integrated and deployed Keycloak as the central identity provider. Developed custom, branded Keycloak themes and configured strict role-based access control (RBAC) to securely separate administrators from standard annotators.
• High-Performance Architecture: Built a resilient, scalable backend using FastAPI, PostgreSQL, and Redis (for advanced rate-limiting and caching). Leveraged Cloudflare R2 for highly available, S3-compatible object storage.
• Modern, Localized Frontend: Crafted a responsive UI using Next.js 16, React 19, and Tailwind CSS v4. Engineered a robust internationalization (i18n) setup featuring a dual-language (English/Arabic) MDX-powered blog and application interface.
• Containerized Infrastructure: Orchestrated a complex, multi-service architecture using Docker Compose, ensuring seamless internal networking and secure deployment across the database, cache, identity server, and application layers.
• Scalable Model Inference Demo (Legacy): Successfully deployed a live demo of a newly released AI model for community testing. Provisioned a specialized GPU VPS running consumer-grade GPUs and optimized the inference engine to serve over 100 concurrent users at peak capacity.