MNK/portfolio

Islamabad · Open to remote work and relocation

AI systems that ship, and keep working after launch.

I'm Muzammil, an AI engineer with 1+ year in production ML, LLM and RAG. I classify healthcare claims at 95% accuracy, build voice agents, and segment 3D MRI scans.

See selected work
Start a conversation
  • PyTorch
  • TensorFlow
  • scikit-learn
  • CatBoost / XGBoost
  • LangChain
  • LangGraph
  • MCP
  • FastAPI
  • Docker
  • Rust

01 — Selected work

Projects with real constraints

Production healthcare systems, from-scratch RAG, agents and vision. Each one has a case study.

All 19 projects →
  • Verifiable Agent Kernel (VAK) workflow diagram
    GenAI & agents

    Verifiable Agent Kernel (VAK)

    v1.0 · 80%+ test coverage

    Deterministic control plane for LLM agents. Least-privilege ABAC, WASM-sandboxed tools, Z3 formal verification before execution, 3-tier memory (Merkle DAG), multi-agent debate with quadratic-voting consensus, and a PRM fine-tuning toolkit. v1.0 with 80%+ test coverage and LangChain/AutoGPT/MCP adapters.

    • Rust
    • Python
    • WASM
    • Cedar/ABAC
    • Z3
  • Healthcare Claims Classification workflow diagram
    Machine learning

    Healthcare Claims Classification

    95% accuracy · 42,900+ claims

    Production classification system (CatBoost, XGBoost, AutoGluon) at 95% accuracy across 42,900+ claims with daily feedback-loop retraining, paired with a denial-prediction and First-Time Pass Rate module that cut the denial rate from 3.2% to 2.4%.

    • CatBoost
    • XGBoost
    • AutoGluon
    • Python
    • Healthcare
  • ELD Trip Planner workflow diagram
    Backend & data

    ELD Trip Planner

    Daily ELD log sheets from a planned trip

    • Django
    • DRF
    • React
  • Document Summarizer workflow diagram
    GenAI & agents

    Document Summarizer

    ~50% lower latency · 99% factual consistency

    • FastAPI
    • Next.js
    • LangChain
  • RAG Custom Engine workflow diagram
    GenAI & agents

    RAG Custom Engine

    ~20% precision gain over dense baselines

    • Python
    • HNSW
    • BM25

02 — Experience

Where I've worked

  1. Aug 2026 - Present

    Data Trainer

    SME Solutions, Inc. via Deel - Remote

    • Create and review detailed prompts and responses used to train AI models across varied topics.
    • Evaluate and rank model responses to improve accuracy in different contexts.
    • Test models for inaccuracies and bias, validating their applicability in the target domains.

  2. Jul 2025 - Aug 2026

    Junior AI Engineer

    CareCloud (MTBC) - Islamabad, Pakistan

    • Claims classification: delivered a production system (CatBoost, XGBoost, AutoGluon) at 95% accuracy across 42,900+ cases, with daily feedback-loop retraining.
    • Denial prediction: built a Python resubmission system with a rule-based First-Time Pass Rate module (provider, payer, CPT, ICD), cutting the denial rate from 3.2% to 2.4% and resubmission wait from 2 weeks to 3 days.
    • Orchestration: designed an n8n workflow acting as a Model Context Protocol (MCP) server for conversation state, prompts and call routing, cutting inference costs 35% through prompt caching.

  3. Mar 2022 - Nov 2025

    Software / AI Engineer

    Suidhaga - Remote

    • Progressed from frontend UI to Django backend to full-stack web features end-to-end over 3+ years for an e-commerce business.
    • Applied data analysis and data science to business data and built ARIMA-family models to forecast product prices.

  4. Sep 2022 - Present

    Freelance AI Engineer

    Independent - Remote

    • Delivered client AI projects, including model training and applied AI application builds, alongside full-time studies and work.

  5. Jul 2024 - Aug 2024

    AI/ML Research Intern

    Risetech - Rawalpindi/Islamabad, Pakistan

    • Segmented brain tumor regions from multi-modal MRI (T1-CE, T2, FLAIR) on the BraTS20 dataset.
    • Benchmarked 3D U-Net against SegFormer3D and selected SegFormer3D (85%+ Dice); used GradCAM for explainability and FastAPI for serving.
    • The work became the Neurofusion FYP: 1st Place at COMPPEC 2025 and a Departmental Silver Medal.

03 — Skills

What I build with

Machine learning

  • PyTorch
  • TensorFlow
  • Keras
  • scikit-learn
  • CatBoost
  • XGBoost
  • AutoGluon
  • Hugging Face

GenAI & agents

  • OpenAI API
  • Anthropic API
  • RAG
  • LangChain
  • LangGraph
  • CrewAI
  • AutoGen
  • MCP
  • n8n
  • ChromaDB
  • FAISS
  • BM25

Vision & voice

  • OpenCV
  • SegFormer3D
  • 3D U-Net
  • YOLO
  • GradCAM
  • LiveKit
  • ElevenLabs
  • Silero VAD

Backend & infra

  • Python
  • Rust
  • C++
  • SQL
  • FastAPI
  • Django / DRF
  • Next.js
  • React
  • Docker
  • Kubernetes
  • CI/CD

04 — Writing

Notes on LLMs and agents

All posts →

05 — Education & credentials

Education, honors and certifications

B.E. Computer Engineering

National University of Sciences and Technology (NUST)

2021–2025 · CGPA 3.25 · Islamabad, Pakistan

Final year project: Neurofusion, 3D brain tumor segmentation on multi-modal MRI (SegFormer3D, BraTS20).

  • 1st Place, COMPPEC 2025
  • Departmental Silver Medal

06 — About

Quick answers

What does Muzammil Nawaz Khan build?
He builds production machine learning, LLM and RAG systems in Python: healthcare claims classification and denial prediction, real-time voice AI agents, retrieval-augmented generation engines, and 3D medical image segmentation.
What results has he delivered?
A claims-classification system at 95% accuracy across 42,900+ cases, a claim denial rate cut from 3.2% to 2.4%, resubmission wait cut from 2 weeks to 3 days, inference costs cut 35% through prompt caching, and front-desk handling time cut 40% with a voice AI agent.
Where is Muzammil based, and is he open to relocation?
He is based in Islamabad, Pakistan, and is open to remote work and relocation.
What is his technical stack?
Python, PyTorch, TensorFlow, scikit-learn, CatBoost, XGBoost, AutoGluon, LangChain, LangGraph, the Model Context Protocol, FastAPI, Django, Next.js, Docker and Kubernetes, plus Rust and C++.
Where did he study and what has he won?
He holds a B.E. in Computer Engineering from NUST (2021–2025, CGPA 3.25). His final year project, Neurofusion, won 1st Place at COMPPEC 2025 and a Departmental Silver Medal.
How can I contact him?
Email mnk.7muzammil86@gmail.com, or reach him on LinkedIn (linkedin.com/in/mnk539) or GitHub (github.com/muzammil5539).

07 — Contact

Have a hard problem? Let's talk.