Danash Mahmood — Portfolio

Projects

A low-decision cooking platform delivering 20 rice cooker one-pot recipes with AI-generated video and narrated cook-along instructions.

            
            Built media generation pipelines using OpenAI Sora for short-form recipe videos and ElevenLabs TTS with word-level timestamps for synchronised narration.
          

            
            Developed a meal planner and automatic shopping list generator that aggregates ingredients across recipes with dynamic unit conversion.
          

            
            Feature-flag driven configuration (e.g. halal-friendly mode) with automated scripts to regenerate recipe media pipelines.
          

            
            Built with Next.js 15 App Router (static export), AI media hosted on Cloudflare R2, deployed via Cloudflare Pages.
          

Next.js 15 OpenAI Sora ElevenLabs TTS Cloudflare R2 Cloudflare Pages

Break It Down ↗ Live

An ADHD-friendly AI planning tool that decomposes complex goals into 5–10 small, actionable steps — dynamically scaling task size based on user energy level.

            
            Serverless edge architecture using Cloudflare Pages Functions with AI inference via Cloudflare Workers AI running Llama 3.1 8B Instruct.
          

            
            Multi-layer AI safety guardrails using Llama Guard 3, hardened prompts, and server-side input validation to prevent unsafe or off-topic requests.
          

            
            Vanilla JS SPA with persistent state via localStorage, supporting iterative step breakdown and continuation workflows.
          

            
            Developed using spec-driven development with Claude Code; automated unit tests with Node's built-in test runner integrated into the deployment pipeline.
          

Cloudflare Pages Workers AI Llama 3.1 Llama Guard 3 Vanilla JS