
Medhat Elhady
Verified Expert in Engineering
AI Engineer and Developer
6th of October City, Giza Governorate, Egypt
Toptal member since July 8, 2026
AI/ML engineer with 4 years of experience building and operating production ML and LLM systems on AWS. Automated MLOps on SageMaker, built RAG and tool-using agents with LangGraph on Amazon Bedrock AgentCore, and ran LLM infrastructure (LiteLLM, Langfuse) on Kubernetes with Terraform. Fine-tuned LLMs on a proprietary language, and contributed a feature to the HashiCorp Terraform AWS Provider
Portfolio
Experience
- Python - 5 years
- SQL - 4 years
- Generative Artificial Intelligence (GenAI) - 3 years
- Amazon SageMaker - 3 years
- Machine Learning Operations (MLOps) - 3 years
- Terraform - 2 years
- Small Language Models (SLMs) - 2 years
- vLLM - 1 year
Work Experience
AI/ML Engineer
Rackspace Technology
- Built an automated Terraform workflow to deploy ML models on Amazon SageMaker endpoints, reducing deployment time by 70%.
- Architected a drift detection agent that relies on retrieval-augmented generation (RAG) and device incidents to identify risks on devices and suggest remediation.
- Fine-tuned AWS Titan and Cohere LLMs on a proprietary XML-based programming language to accelerate boilerplate code generation for business users, raising accuracy from 65% to ~92% (LLM-as-a-judge and cosine similarity).
- Grew the training set from 200 to 12K examples by extracting samples from XML based code and labeling them with an LLM plus RAG over the language documentation.
- Hosted LiteLLM on Kubernetes as a unified OpenAI-compatible gateway across multiple LLM providers, handling model routing, fallback and access control.
- Deployed Langfuse to trace LLM calls and built dashboards to track RAG metrics. Built a Drift Detection Agent using RAG over device incidents to identify risks and recommend remediation.
- Migrated seven teams from JupyterHub to SageMaker with team isolation, S3 access controls and instance provisioning, cutting monthly costs by 60% mainly through automatic notebook instance shutdown.
- Contributed to the HashiCorp Terraform AWS Provider (PR #44407): added an additional data source parameter to the SageMaker model resource, so a scikit-learn processor artifact can be passed to the model at deployment.
Mentor
Udacity
- Worked for three years on the FWD initiative in the data analysis track and the DECI initiative, teaching Programming Fundamentals Level 2.
- Reviewed student projects and provided feedback on areas for improvement, as well as suggestions for enhancing their work in fields of generative AI, agentic systems, and Amazon Bedrock.
- Answered 1,500+ questions, created 50+ articles, conducted over 40 webinars and more than 100 sessions, and reviewed 2,000+ projects in the field of data, generative AI, and programming.
BI and Data Warehouse Specialist
Telecom Egypt
- Migrated real-time cabinet-level outage data into the customer service portal, enabling automated customer notifications and reducing customer agent workload by 40%.
- Developed and maintained ETL processes that transform data from its raw form into a format that can be easily analyzed and visualized using SQL and Talend ETL.
- Engineered a sentiment analysis proof of concept to automatically classify customer reviews for marketing campaigns.
Project History
Slack Support Agent on Amazon Bedrock AgentCore
Deployed the containerized agent on AgentCore Runtime in a VPC with AgentCore Memory, and routed all model traffic through an in-VPC LiteLLM gateway on EKS for routing, cost control and access control.
Added opt-in PII guardrails (Bedrock Guardrails and Presidio) on input, output and logging, plus guardrail evaluation suites for PII and jailbreak testing
Traced every turn in Langfuse with tool audit metadata (KB citations, relevance scores, escalations) alongside CloudWatch; provisioned the platform with modular Terraform.
End-to-End MLOps Pipeline on SageMaker
Deployment phase: manager approval of the model in the registry triggers a Lambda function that deploys it to a SageMaker endpoint with data captured in s3.
Releases use canary deployment with rollback alerts, and the endpoint autoscales with workload.
Clinic Front-Desk Copilot
Exposed scheduling and knowledge tools through MCP servers with per-request token auth and cross-patient access tests.
Evaluated on scripted bilingual conversations for booking success, red-flag recall and grounded FAQ answers, and load-tested with Locust under concurrent users to check latency and confirm no double bookings.
RAG Application with Amazon Bedrock
https://github.com/medhatelhady/Building-Generative-AI-Applications-with-Amazon-Bedrock-and-pythonOffline RAG Evaluation Pipeline
The job replays ground-truth Q&A (JSONL in S3) against the agent running in offline mode, then scores answers with deterministic retrieval metrics and RAGAS metrics using an LLM judge served through LiteLLM.
Published scores to Langfuse custom dashboards and stored a summary for each run in S3.
Movie Search Agent
https://github.com/medhatelhady/IMDB-Search-AgentUsed PostgreSQL as the single data store: pgvector for semantic search over sentence-transformers embeddings and pg_trgm for fuzzy title matching, fed by an ingestion pipeline over the TMDB 5000 movie dataset.
Served the agent through a FastAPI API with streaming responses and a Streamlit chat UI with session memory; runs on OpenAI or local Ollama models, with the database in Docker Compose.
Find Donors for Charity | Machine Learning Classification Project
https://github.com/medhatelhady/Find-Donors-for-CharityMLDog Image Classifier with AWS
https://github.com/medhatelhady/dog-images-classifier-using-awsEducation
Bachelor's Degree in Computer Engineering
Zagazig University - Zagazig, Egypt
Certifications
AWS Certified Generative AI Developer – Professional
Amazon Web Services (AWS)
Machine Learning in Production
DeepLearning.AI
AWS Certified Machine Learning – Specialty
Amazon Web Services (AWS)
AWS Machine Learning Engineer Nanodegree
Udacity
Machine Learning
Coursera
AI Programming with Python Nanodegree
Udacity
Skills
Libraries/APIs
PyTorch, TensorFlow, Scikit-learn, Pandas, NumPy, Matplotlib, vLLM, Keras, XGBoost
Tools
Terraform, Amazon SageMaker, Talend ETL, Git, GitLab, AWS IAM, Seaborn, Jenkins, Jupyter, Docker Compose, Amazon CloudWatch
Languages
Python, SQL, Bash
Frameworks
LangGraph, Agentic Frameworks, Streamlit, Locust
Platforms
Amazon Web Services (AWS), Docker, LangSmith, Langfuse, AWS Lambda, Linux, Ollama
Paradigms
Model Context Protocol (MCP), Anomaly Detection
Storage
Databases, PostgreSQL
Other
RAG Systems, Agentic RAG Systems, Transformers, LangChain, Prompt Engineering, Fine-tuning, EDA, Data Modeling, Deep Learning, Natural Language Processing (NLP), Data Processing, Feature Engineering, Model Evaluation, Terraform (IaC), CI/CD Pipelines, Agentic AI, Artificial Intelligence (AI), Applied AI, Large Language Models (LLMs), Large Language Model Operations (LLMOps), Data Science, AI Agents, AI Agent Orchestration, OpenAI, RAG Architecture, Conversational AI, Performance Optimization, Monitoring, Agentic AI Systems, Agentic Workflow Design, AI Engineering, AI Architecture, Retrieval-augmented Generation (RAG), Inference Optimization, Web Scraping, Machine Learning, Data Structures, Algorithms, Small Language Models (SLMs), LLM Fine-tuning, Amazon Bedrock AgentCore, Model Validation, Machine Learning Operations (MLOps), Data Analysis, Generative Artificial Intelligence (GenAI), AI Programming, Hyperparameter Tuning, Data Augmentation, Computer Vision, Amazon RDS, Embedding Models, Deep Neural Networks (DNNs), Deployment, Observability, Supervised Learning, Clustering, AWS ECS Fargate, FastAPI, Pgvector, RAGAS, Amazon EventBridge, Amazon Bedrock Knowledge Bases, LiteLLM, Sentence Transformers, Autoscaling
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring