projects

an autonomous agent that earns its own autonomy

BlackAgent starts with limited capabilities and earns broader access by demonstrating competence, the same way a new team member might. Built around a strict control loop that separates what must be predictable from what can be creative, the system runs a 70+ tool runtime that can exceed 100 tools with optional integrations, maintains layered memory that fades like human recall, and detects intent from natural conversation without being explicitly told what to do. It currently runs on a Raspberry Pi 5 in Casablanca.

a companion that sees, in the language you actually speak

Over 350,000 Moroccans live with visual impairments. Nadar is a real-time voice companion that sees, remembers, and navigates alongside its user in Moroccan Darija, French, or English. It recognizes faces and remembers names. It builds a relationship graph of people in the user's life. It reads menus, medicine labels, and street signs through OCR. It remembers facts across sessions through semantic memory that works across languages. It finds nearby pharmacies and tells you if they are open. It runs on anything with a microphone, a speaker, a camera, and an internet connection. It was built because the people who need it most have been waiting the longest.

a translator between what you mean and what the model needs to hear

AI is a gold in, gold out machine. If you feed it vague input, it does not fail gracefully. It amplifies the vagueness and returns something confidently mediocre. Most people experience this as the model being bad. It is not. The model is doing exactly what it was asked to do. The problem is that most people do not know how to ask. SudoPrompt is a pre-production layer that sits between human intent and model execution. Instead of letting the user fire off a half-formed idea and hope for the best, it forces a clarification loop, called the Sudo Check, that interrogates the request, surfaces missing constraints, and only then generates. It does not guess what you meant. It asks until it knows. Three modules handle different domains: Visual Studio for image and video prompts, Agent Forge for system prompts and agent architectures, and Docs Builder for PRDs, implementation plans, and technical documentation.

a visual engine that draws the motion your eyes skip over

Video contains more information than what appears on screen. Between every two frames there is motion, and that motion has direction, speed, rhythm, and shape. TRAJEKT is a real-time engine that captures pixel-level motion from any video source, stores it as spatial memory, and synthesizes geometry directly from that data. The system divides each frame into a sampling grid, tracks brightness changes across frames, estimates displacement vectors, and feeds that trajectory data into any of its synthesis modes that draw curves, shatter fragments, advect particles, deform meshes, or decompose frequencies. The same input with the same parameters always produces identical output. The aesthetic comes from math applied to real spatial data.