MUZAMMIL NAWAZ KHAN

AI Engineer | Agentic Systems & Voice AI Specialist | 95% Claims Accuracy @ CareCloud

Based in: Islamabad, PakistanCurrently: Data Trainer, SME Solutions, Inc. via DeelExperience: 5 years

AI Engineer with a track record of deploying production-grade agentic systems, including autonomous voice agents and high-accuracy classification pipelines. Expert in LangGraph, MCP, and n8n, with a focus on cutting operational costs and improving accuracy in complex domains like healthcare and e-commerce. Known for achieving 95% accuracy in production claims classification and reducing inference costs by 35%.

Experience

Data Trainer

SME Solutions, Inc. via DeelRemotecontractAug 2026 – Present

Create and review detailed prompts and responses used to train AI models across varied topics. Evaluate and rank model responses, and test models for inaccuracies and bias in their target domains.

AI modelsPrompt Engineering

Freelance AI Engineer

IndependentRemotefreelance2022 – Present

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

Model TrainingApplied AI

Junior AI Engineer

CareCloud (MTBC)Islamabad, PakistanworkJul 2025 – Aug 2026

Summary

Designed an n8n workflow acting as a Model Context Protocol (MCP) server for conversation state, prompt caching and tool routing in a multi-agent voice stack, cutting inference costs 35%. Worked across a real-time autonomous voice AI agent (LiveKit, LangGraph, OpenAI, Silero VAD, SIP) that scheduled, routed and resolved calls end-to-end, cutting front-desk handling time 40%. Delivered a production claims-classification pipeline (CatBoost, XGBoost, AutoGluon) at 95% accuracy across 42,900+ cases, with daily feedback-loop retraining for drift resilience. Built a Python denial-prediction and resubmission system with a rule-based First-Time Pass Rate module (CPT/ICD), cutting the claim denial rate from 3.2% to 2.4%.

What I did

Delivered a production claims-classification pipeline (CatBoost, XGBoost, AutoGluon) with daily feedback-loop retraining for drift resilience

Results

Cutting inference costs 35% via MCP server design.
Cutting front-desk handling time 40% via autonomous voice AI agent.
Achieved 95% accuracy across 42,900+ cases in production claims-classification.

n8nModel Context Protocol (MCP)LiveKitLangGraphOpenAISilero VADSIPCatBoostXGBoostAutoGluonPythonCPT/ICD

Skills

Technical

LangGraph

LangChain

ReAct agents

OpenAI API

Prompt Engineering

HNSW

Python

MCP

n8n

Hugging Face

Anthropic API

LLM-as-judge evaluation

FastAPI

Tool-use design

ChromaDB

FAISS

Django/DRF

BM25

XGBoost

PyTorch

Reciprocal Rank Fusion

CatBoost

LlamaIndex

Self-RAG

OpenCV

CrewAI

AutoGen

Docker

TensorFlow

Azure

Kubernetes

Node.js

Projects

Verifiable Agent Kernel (VAK)

Lead Developerpersonal

Control plane for LLM agents with ABAC policy gating, WASM-sandboxed tools and Z3 formal verification; v1.0 with 80%+ test coverage on Docker/Kubernetes CI/CD.

RustPythonWASMCedarZ3DockerKubernetes

View the code

Conversational AI Agent (ReAct)

Developerpersonal

6 tools with visible reasoning traces, WebSocket streaming, SQLite checkpointer memory.

LangGraphFastAPINext.jsOpenAIWebSocketSQLite

View the code

RAG Custom Engine

Developerpersonal

From-scratch HNSW + BM25 + Reciprocal Rank Fusion with a 3-stage Self-RAG gate, ~20% precision gain over dense-vector baselines.

PythonGPT-4o-miniHNSWBM25Reciprocal Rank Fusion

View the code

Education

F.Sc. Pre-Engineering

Ghazali Premier College, LahorePre-Engineering

Bachelor of Computer Engineering

National University of Sciences and Technology (NUST)Computer EngineeringGPA 3.252021 – 2025

Recognition

Awards

1st Place at COMPPEC 2025

COMPPEC2025

Won for work on brain tumor segmentation benchmarking 3D U-Net against SegFormer3D.

Departmental Silver Medal

National University of Sciences and Technology (NUST)

Awarded for research excellence in AI/ML.

Certifications

IBM AI Engineering

IBM

IBM Generative AI Engineering

IBM

IBM RAG & Agentic AI

IBM

IBM Deep Learning with PyTorch, Keras & TensorFlow

IBM

LLMOps Specialization

Duke University

Informed Clinical Decision Making using Deep Learning

University of Glasgow

Hire MUZAMMILReach out about a role, a contract or a conversation.For recruiters

MUZAMMIL's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis