Ahmad Khalid

AI Engineer with hands-on experience designing, training, and deploying production-grade ML systems in cloud environments. At CareCloud, built a two-stage CatBoost denial prediction pipeline processing over one million insurance claims, an async batch processing backend managing dual SQL Server databases, and a multi-modal OCR pipeline using Google Vertex AI. Experienced with MLflow for experiment tracking and model versioning, FastAPI for model serving, and Python-based data pipeline construction.

Familiar with GCP services including Vertex AI and Gemini, with growing exposure to Azure tooling. Strong foundation in software engineering practices, model lifecycle management, and deploying AI solutions that operate reliably in production.

AI Engineer at CareCloud MTBC (2025-07 – Present)

• AI Denial Prediction Service: Engineered a production FastAPI microservice deploying a two-stage CatBoost pipeline (binary parent model feeding a multiclass child) to predict insurance claim denial categories per CPT line from raw 837 EDI data, applying confidence thresholds of 90% on the parent and 80% on the child model with per-class probability encoding across 171 engineered features.
• Data Pipeline and Batch Processing: Built a dual-database batch processing service (pymssql) managing 6 SQL Server tables across MIS DB and AI DB, handling claim routing, audit logging, and clinical note insertion; implemented async concurrent processing (aiohttp + asyncio) with configurable batch sizing to handle throughput across thousands of claims per run.
• Cloud AI Integration with GCP Vertex AI: Engineered a multi-modal OCR pipeline using Google Vertex AI (Gemini 2.5 Pro) for structured data extraction from HCFA-1500 medical forms and EOBs, achieving 92% extraction accuracy and automating end-to-end document processing workflows.
• Claims Classification at Scale: Developed an XGBoost-based classification system processing 1M+ medical records at 95% accuracy on balanced datasets, reducing manual analyst workload by 40% and improving first-pass claim submission rates.
• LLM-Powered Analysis Agent: Built a GPT-4 agent using LangChain and LangGraph to provide structured claim rejection reasoning and resubmission recommendations, integrated into the live claims workflow as a downstream step after the ML prediction pipeline.
• MLOps and Model Lifecycle: Used MLflow throughout for experiment tracking, model versioning, and reproducibility across multiple concurrent model development efforts; maintained FastAPI-based inference services with monitoring via Streamlit dashboards processing 100+ daily transactions.
• Medicare Prevention System: Automated GW-modifier detection on EDI 270/271 eligibility data, preventing approximately $187,000 per year in claim denials with sub-5-second response time on live data.

System Specialist at National University of Sciences and Technology (NUST) (2024-09 – 2025-08)

• Wi-Fi CSI Occupancy Detection (Final Year Project): Designed and deployed a real-time ML classification system using Wi-Fi Channel State Information, achieving 92% accuracy on a live pipeline processing 1,000+ measurements per second; built the accompanying React and FastAPI dashboard supporting 50 concurrent users.

Bachelor of Engineering in Computer Engineering – National University of Sciences and Technology (NUST) (2021-09 – 2025-05)