I am a founding AI engineer who builds the first version of things that do not exist yet: AI infrastructure, agent workflows, retrieval systems, validation loops, and production automations. My work turns messy operational processes into reliable pipelines that teams can actually use.
| Multi-agent LangGraph workflows, browser automation, MCP servers, tool orchestration, and guardrail-driven execution. | OCR/VLM extraction, adaptive schemas, duplicate detection, validation loops, and structured exports for large batches. | Hybrid retrieval, LLM integrations, drift monitoring, retraining pipelines, API serving, and production observability. |
Founding AI Engineer at Nobility RCM | Islamabad, Pakistan | March 2025 - Present
Joined as the company's founding AI engineer and started the AI function from scratch: technical direction, architecture, model workflows, automation strategy, evaluation patterns, and production delivery.
- Established the initial AI engineering stack for healthcare operations, including LangGraph agents, RAG pipelines, MCP integrations, Playwright browser automation, and validation-first LLM workflows.
- Built a medical compliance agent that reads clinical notes and extracts ICD-10, CPT, and HCPCS billing codes with a fine-tuned Phi-3 model, a 4-agent LangGraph workflow, and layered anti-hallucination checks.
- Developed LLM-powered browser automation agents with Playwright and Browser Use to navigate billing portals, extract claim data, and auto-fill repetitive workflows.
- Designed retrieval and knowledge workflows with FAISS vectors, BM25 keyword search, and internal system integrations so teams can query billing data in natural language.
- Turned early prototypes into repeatable internal systems by defining prompts, tools, state flows, validation steps, and operational handoff patterns.
|
Third Prize
Qwen-Image LoRA Training Competition |
12k+ Images
Curated ad dataset |
40 Categories
Commercial product domains |
7.47 Score
AI for Production track |
|
Multi-step workflows with explicit tools, state, retry paths, and validation layers instead of loose chatbot behavior. |
Systems that connect to real data, real portals, real users, and measurable outcomes. |
From research prototype to usable workflow with clean evaluation loops and practical deployment constraints. |






