Sayyam Shahbaz obligator11

👋 Hi, I'm Sayyam Shahbaz

⚡ About Me

AI engineer:
  name: "Sayyam Shahbaz"
  location: "Pakistan"
  certifications: ["Microsoft Certified AI Engineer", "IBM Certified AI Engineer"]
  specialization: ["Software Architecture", "System Design", "Agentic AI Engineering"]
  currently_building: "Multi-Agent Digital Ops Team — IT Helpdesk MVP"
  philosophy: "Local-first, zero-paid-API, human-in-the-loop by design"
  learning: ["Agentic architectures", "MCP integrations", "Applied spatial math for CV"]
  brand: "@obligator11 — content on TikTok / Instagram / LinkedIn / YouTube"

I design and ship multi-agent AI systems that run entirely on local infrastructure — orchestrating models like Qwen2.5-Coder, DeepSeek-R1, Llama 3.1, and Gemma through Ollama and LM Studio, with PostgreSQL, ChromaDB, and Redis handling state, memory, and concurrency. Every system I build follows the same four-layer discipline: Presentation → Application → Domain → Infrastructure.

Before this, I spent months building real-time computer vision and AR systems (pose estimation, gesture control, YOLO-based tracking) — that CV depth now shows up in how I think about latency, threading, and perception pipelines inside agent systems.

🧠 Currently Architecting

🎫 Multi-Agent Digital Ops Team IT Helpdesk MVP · v3 architecture Redis/RQ concurrency, Prometheus + Grafana observability,

🧾 Invoice/AP Automation Agent 4-model pipeline: extraction → anomaly reasoning → explanation → routing, with a human-in-the-loop approval gate

🧠 Local Dual-LLM RAG Workspace NotebookLM-style research tool — DeepSeek-R1 (reasoning) + Qwen2.5-Coder (implementation) via LM Studio, isolated per-notebook ChromaDB vaults

🏗️ How I Build — Four-Layer Architecture

Every agentic system I ship follows this pattern:

graph TD
    A["🎨 Presentation Layer<br/>Streamlit"] --> B["⚙️ Application Layer<br/>Agent Pipeline / Orchestration"]
    B --> C["📐 Domain Layer<br/>Pydantic Schemas"]
    B --> D["🗄️ Infrastructure Layer<br/>Local LLMs · PostgreSQL · ChromaDB · Redis"]
    D --> E["🧠 Qwen2.5-Coder<br/>Extraction"]
    D --> F["🔍 DeepSeek R1<br/>Reasoning"]
    D --> G["💬 Llama 3.1<br/>Explanation"]
    D --> H["🧭 Gemma<br/>Routing"]

    style A fill:#0d1117,stroke:#00f5ff,color:#00f5ff
    style B fill:#0d1117,stroke:#bd00ff,color:#bd00ff
    style C fill:#0d1117,stroke:#00f5ff,color:#00f5ff
    style D fill:#0d1117,stroke:#bd00ff,color:#bd00ff
    style E fill:#1a1a3e,stroke:#00f5ff,color:#e0e0e0
    style F fill:#1a1a3e,stroke:#00f5ff,color:#e0e0e0
    style G fill:#1a1a3e,stroke:#00f5ff,color:#e0e0e0
    style H fill:#1a1a3e,stroke:#00f5ff,color:#e0e0e0

📌 Top Repositories

🤖 Agentic AI & Automation

Invoice/AP Automation Agent — 4-model local pipeline (Qwen2.5-Coder → DeepSeek R1 → Llama 3.1 → Gemma) with PostgreSQL + ChromaDB and a human-in-the-loop approval gate
Local Dual-LLM RAG Workspace — NotebookLM-style research tool combining DeepSeek-R1 + Qwen2.5-Coder via LM Studio, with per-notebook ChromaDB vaults

👁️ Computer Vision & AR Suite

Vision-Core-Projects — 60+ real-time CV/AR experiments: pose-driven games, gesture control, AR overlays, all single-file with procedural audio
CrowdAI — real-time crowd density & flow analytics using YOLOv8 + DBSCAN clustering
DriverFatigueSystem — MediaPipe Face Mesh–based fatigue/attention monitor using EAR/MAR metrics
InterviewAnalyzer — multi-modal interview confidence analyzer (MediaPipe Face Mesh + Pose + Whisper)

🏢 Client & Product Work

Solid Gym Management System — modular PySide6 desktop app with RBAC, financial transaction logging, and Google Sheets–backed state
Triple Eyes Real Estate & Marketing — React/TypeScript site with Tailwind CSS and Framer Motion for an Islamabad architectural firm