Skip to content

// 04 / WORK

Work

Newest products first: agency ops, agent tooling, synthetic conversation, multi-agent worlds, and furniture AR - then selected research and applied ML projects.

HUD-style abstract of CRM pipeline cards and connected agent workflow nodes

// PRODUCT · AGENCY OPS

Product

Agency OS

CRM, agent runtime, acquisition, and delivery in one Next.js platform

Production-minded agency operations product: CRM foundation, durable agent runtime, lead acquisition, decision-maker intelligence, gated outreach, and service delivery - with demo auth, workers, and release-hardening docs.

TypeScriptNext.jsCRMAgent runtimeProduct
Open →
Abstract skill-router diagram with a central node linking to capability capsules

// PRODUCT · DEVELOPER TOOLS

Product

OmniSkill

Universal AI skill router and workflow planner for local agents

CLI product that fingerprints a repo, classifies tasks, discovers skills with explainable scores, plans multi-stage workflows, and builds provenance-aware bundles for Claude Code, Cursor, and portable agents - offline-first, no invented success metrics.

TypeScriptCLIAI skillsCursorClaude Code
Open →
Abstract weave of agent threads and typed tool nodes over a dark HUD grid

// PRODUCT · AGENT FRAMEWORK

Product

AgentWeave

Observable Python agents with typed tools, traces, and replay

Lightweight Python library for tool-using agents: typed tools, safety limits, multi-agent role routing, execution traces with secret redaction, and a Replay & Regression Lab that runs offline with MockProvider - without a full SaaS stack.

PythonAgentsTyped toolsTracesReplay
Open →
Wireframe synthetic avatar profile with conversation-state HUD arcs

// PRODUCT · CONVERSATIONAL AI

Product

LivePersona

Interruption-aware synthetic avatar conversations

Interactive avatar conversation product with a real barge-in state machine: listen → transcribe → generate → speak → interrupt → recover. Demo mode works without API keys; optional streamed LLM, browser STT/TTS, and a framework-independent core package.

TypeScriptAvatar UXSTT/TTSStreamingProduct
Open →
Grid world map with character dots and a forked world-comparison panel

// PRODUCT · MULTI-AGENT SIM

Product

Everworld AI

Persistent multi-agent worlds with Experimental Worlds comparison

Browser sandbox where a prompt becomes a living grid society of autonomous characters with memory, relationships, and resources. Experimental Worlds forks a save, changes one variable, and compares real simulation metrics - rule-based demo mode included.

TypeScriptMulti-agentSimulationExperimental Worlds
Open →
Wireframe L-shaped sofa with AR placement markers in a dark room plane

// PRODUCT · AR / 3D

Product

KUKA HOME Furniture AR

Web AR catalog for an L-shaped sectional in-room preview

Product experience for furniture visualization: browse a KUKA HOME L-shaped sectional, inspect it in 3D, and place it in a room via AR Quick Look (iOS) or Scene Viewer (Android) using GLB/USDZ assets.

Web AR3DRetailProduct vizJavaScript
Open →
Abstract HUD render of an embedded ML inference pipeline with an ECG waveform and sensor readouts

// PROJECT · APPLIED ML

Project

Injury Detection System

SpO2 & heart rate telemetry to injury classifier

Research/project prototype: reads heart rate, SpO2, and temperature from Arduino sensors and classifies injury risk in real time, with reported ~95% accuracy in early identification during project evaluation, plus automated alert workflows. Not a certified medical product.

PythonC++ / ArduinoEmbedded MLPrototype
Open →
Abstract HUD render of an NLP knowledge graph with glowing semantic nodes

// PROJECT · NLP · AI UX

Project

TinyTask (AI Note System)

Contextual summarization and task memory

Concept and UI architecture for a note-taking platform with contextual summarization and a simple, focused workflow.

NLPSummarizationMinimalist UIFletPrototype
Open →
Abstract HUD render of injection arrows being filtered by a cyan shield-shaped model guardrail

// PROJECT · AI SAFETY

Project

Prompt Injection Security Engine

Prompt-injection detection for LLMs

Research prototypes that detect prompt-injection attacks on LLMs, with filtering pipelines and guardrails against unsafe execution.

LLM GuardrailsAdversarial MLPythonResearch
Open →
منتجات ومشاريع ذكاء اصطناعي وتقنية | ليث أبو سليم