Muhammad Abdullah Malik
I help teams move from an AI idea to a working system in the field: autonomous agents, lightweight Go runtimes, and MCP-powered workflows that are observable, secure, and ready for real users.
Engineered AI Systems
Production autonomous agents, edge runtimes, Model Context Protocol servers, and enterprise automation pipelines.
MalikClaw
Edge-Native Agentic AI Assistant & Gateway
Ultra-lightweight, high-performance agentic AI assistant & gateway in Go, optimized for edge hardware ($10 Raspberry Pi/Android) with Urdu-First support.
Standard Python agent frameworks consume 300MB–1GB+ RAM, making them unusable on constrained edge hardware like a $10 Raspberry Pi Zero or Android devices.
A Go-based lightweight compiled agent runtime with Model Context Protocol (MCP) tool standard, sandboxed ADB automation, and native RTL Urdu NLP.
Digital FTE — Abdullah Junior
24/7 Autonomous Digital Employee
High-autonomy AI agent acting as a 24/7 Digital Employee for personal and enterprise task execution with dual-agent security isolation.
Handling high-volume daily operations across Email, WhatsApp, and CRM without human bottlenecks while keeping private credentials safe from cloud exposure.
Dual-agent Cloud/Local architecture separating read-only monitoring from privileged local task execution, synchronized via HMAC-signed message buses and MCP.
-10b981?style=flat-square&logo=claude&logoColor=white){kind=icon}
Customer Success Digital FTE
Enterprise Support Agent with Kafka & pgvector
Enterprise support agent handling inquiries 24/7 across Email, WhatsApp, and Web with pgvector semantic retrieval and Kafka async queues.
Delivering instant, contextual customer support across email, WhatsApp, and web tickets under traffic spikes without dropped requests or hallucinated answers.
Kafka-driven asynchronous agent with 5 function tools, pgvector knowledge base retrieval, and automatic sentiment escalation.
Other Specialized Systems & Tools
Full-stack applications, developer tooling, and educational AI platforms.
Deep Research Agent
Autonomously conducts deep multi-step web research, synthesizes evidence, and generates structured academic reports.
Voice Assistant Agent
Privacy-first voice assistant with wake word detection, semantic memory, and local hardware control.
Physical AI Platform
Interactive learning platform for Physical AI & Humanoid Robotics with a RAG chatbot and 43+ code examples.
AI Code Assistant
Multi-model coding agent that assists developers with writing, debugging, and refactoring code with MEM0 memory.
Portfolio AI Assistant Platform
Interactive portfolio interface with embedded RAG chatbot, tool execution, and database persistence.
Online Quran — Asad Ali
Modern interactive web application providing accessible digital Quran reading and search capabilities.
Azan Shop POS Software
Production commercial desktop point-of-sale and retail inventory system engineered for high-throughput billing and cash operations.
Unique Boutique Platform
Custom e-commerce and inventory management platform powering catalog management, orders, and sales tracking.
Engineering Case Studies
Architectural breakdowns of flagship autonomous AI systems built for production performance, security isolation, and edge efficiency.
MalikClaw — Edge Agentic Runtime
Ultra-lightweight AI Assistant Gateway for Edge Devices with Urdu-First Support
Standard Python agent frameworks (e.g. LangChain, CrewAI) consume 300MB–1GB+ RAM, making them unusable on cheap $10 Raspberry Pi Zero or Android hardware.
- •Memory footprint <15MB RAM
- •Native Urdu & English bilingual command parsing
- •ADB interface for physical Android screen automation
- •Compliance with Model Context Protocol (MCP) standards
Inbound Request → Go HTTP/gRPC Gateway → Urdu/English Intent Router → Local Tool Dispatcher (ADB/Shell) → MCP Server Bridge → State Store
Single-binary compiled Go daemon with asynchronous event queue. Uses lightweight thread locks instead of heavy Python runtime inter-process communications.
File-backed JSON state machine with persistent vector embeddings stored locally on disk for zero cloud DB dependency.
Exposes standardized JSON-RPC 2.0 transport over stdin/stdout and SSE HTTP endpoints. Listed on official MCP Market registry.
Hybrid local-cloud: Edge router handles regex/intent classification; falls back to Gemini 1.5 Flash / Claude 3.5 Sonnet for multi-step reasoning.
Autonomous retries with exponential backoff on ADB disconnections, fallbacks to Urdu audio feedback on execution failure.
Benchmarked boot time (<850ms), idle memory (<8.5MB RAM), and task completion accuracy across 100+ simulated edge execution commands.
- Rewrote core runtime in Go (Golang) to eliminate Python memory overhead
- Designed custom RTL (Right-to-Left) Urdu parser for regional accessibility
- Secured agent sandbox via OS-level privilege separation
Achieved 98.4% RAM reduction vs traditional gateways, successful deployment on $10 edge hardware, and official A2AS behavior security certification.
Agentic System Architecture
How I architect production multi-agent systems: end-to-end lifecycle from edge ingress and routing to MCP execution, vector memory, and A2AS safety governance.
AI Gateway & Ingress
Unified API & Event Router
Inbound requests from Web, WhatsApp, Android ADB, and Email are authenticated, rate-limited, and normalized into JSON-RPC event streams.
// Go HTTP Gateway & Ingress
func (gw *Gateway) RouteInbound(req *Request) (*Event, error) {
if err := gw.RateLimiter.Allow(req.ClientID); err != nil {
return nil, ErrRateLimitExceeded
}
return gw.EventBus.Publish(req.Context(), req.Payload)
}Engineering Credibility & Discipline
How I approach building autonomous AI software: deterministic tooling, strict security boundaries, and verifiable observability.
Model Context Protocol (MCP)
Enforces strict JSON-RPC 2.0 transport schemas between LLM runtimes and local/cloud resources (Odoo, Gmail, ADB, Obsidian).
mcp.RegisterServer("malikclaw", transport.Stdio())Cloud/Local Dual-Agent Split
Decouples read-only cloud monitoring (Cloud Sentry) from privileged desktop execution (Local Executive) to protect API secrets.
CloudSentry(Readonly) -> Bus -> LocalExecutive(Privileged)RAG & Vector Search
Implements pgvector and ChromaDB embedding similarity search for fast, sub-second contextual knowledge retrieval.
SELECT * FROM docs ORDER BY embedding <=> $1 LIMIT 5;Behavior Compliance & Safety
Audits and declares agent behavior boundaries using A2AS Registry certificates for production governance.
a2as.VerifyBehaviorCertificate("deep-research-agent")Multi-Agent Orchestration
Coordinates specialized subroutines (Research, Coding, Financials) with dynamic intent routers and model fallback cascades.
Router.Dispatch(intent) -> Select(Gemini1.5, Claude3.5)Observability & Telemetry
Tracks LLM token consumption, execution latency, and function call success rates using Langfuse and structured logging.
langfuse.Trace(span_id, tokens_used, latency_ms)Engineering Capabilities Matrix
Interactive technical matrix organizing agentic frameworks, high-performance runtimes, vector memory, and governance tools.
Agentic Frameworks & Protocols
Orchestration layers, tool communication standards, and multi-agent systems.
Model Context Protocol (MCP)
Standardized JSON-RPC 2.0 tool transport for MalikClaw & Digital FTEs.
OpenAI Agents SDK
Autonomous multi-step research loops, handoffs, and function calling.
Multi-Agent Swarms
Cloud Sentry + Local Executive dual-agent separation with HMAC buses.
Digital FTE Orchestration
24/7 background worker architectures with self-healing retries.
A2A & N8N Automation
Visual agent trigger pipelines and webhook synchronization.
Professional Certifications
Audited AI agent behavioral governance, Skills Directory Grade-A security, prompt engineering specialization, and verified engineering credentials.
A2AS Behavior Certificate — deep-research-agent
A2AS Registry (a2as.org)
Certified by the A2AS Registry for AI Agent Security and Governance. Validates that deep-research-agent operates within audited, declared behavioral boundaries.
Prompt & Context Engineering: Effective AI Communication
Panaversity
Certificate of Achievement conferred 31 October 2025 for passing the Prompt and Context Engineering: Effective AI exam, verified by the Panaversity Academic Council.
Skills Directory Certified Author — 12 Grade A Skills
Skills Directory (skillsdirectory.com)
Verified author credentials with 12 published Claude agent skills passing automated static analysis, prompt injection defense, and malware screening with Grade A certification.
About Abdullah Malik
A Forward Deployed Engineer in training: close to the workflow, precise about the system, and focused on shipping useful autonomy.
Technical Journey & Evolution
Full-Stack Foundations & Systems
Mastered web fundamentals, asynchronous runtime architectures, and core JavaScript/TypeScript backend pipelines.
Agentic AI Specialization at Panaversity & PIAIC
Completed advanced Agentic AI program. Mastered OpenAI Agent SDK, LangChain, N8N, and autonomous tool calling schemas.
Digital FTEs & Go Edge Gateways (MalikClaw)
Engineered MalikClaw (Go edge runtime, 20 GitHub stars) and Digital FTE systems with Cloud Sentry + Local Executive dual-agent isolation.
BS Data Science (IUB) & Enterprise AI Swarms
Pursuing BS Data Science at Islamia University of Bahawalpur (IUB) while scaling cloud-native agent orchestration (Docker, pgvector, Kafka, MCP) with audited A2AS behavior compliance.
Abdullah Malik
Forward Deployed EngineerI translate real operating workflows into secure agent systems, from Go edge daemons (<10MB RAM) to MCP integrations and Digital FTEs.
Core Engineering Principles
01. Autonomous Efficiency First
AI agents should execute seamlessly without heavy runtime baggage. MalikClaw proves agents can operate reliably on <10MB RAM on edge devices.
02. Strict Privilege & Credential Isolation
Cloud event monitoring must be decoupled from privileged local execution. Secrets stay local; sentries stay read-only.
03. Deterministic Tooling (MCP & A2AS)
Tool calling must follow verifiable transport protocols like Model Context Protocol, and agent behavior must be declared and audited.
Have an agent system worth building?
Send the workflow you want automated. I reply to every genuine inquiry within 24 hours — usually with a first-pass architecture sketch.
- +923040705172
- Availability
- Open to agentic AI engineering work
