AI/Data Engineer

Mahdi
Rafiei.

Mahdi Rafiei profile picture

Building intelligent systems that
learn, adapt, and scale

I build scalable data platforms, event-driven cloud infrastructure, and production AI systems. Specializing in data engineering, AWS, and LLM-powered applications.

~ $ python train.py --model gpt-custom
Loading dataset... 142k samples
Epoch 1/10 ━━━━━━━━━━ loss: 2.341
Epoch 5/10 ━━━━━━━━━━ loss: 0.892
Epoch 10/10 ━━━━━━━━━━ loss: 0.341
Checkpointing to ./runs/exp_042
Eval accuracy: 94.7%
~ $

5+

Models shipped to prod

13B+

Parameters trained

4

AI systems designed

4yrs+

Industry experience

03 — Projects

Selected work.

001

Receipt Detection System

Production YOLO model deployed on AWS ECS Fargate detecting receipts in real time. Secured with Amazon Cognito, served via CloudFront, with a full web interface for live inference.

YOLOv8 AWS ECS FastAPI Cognito
Live
Try live demo

002

Tokenless

Run LLMs on Kaggle's free GPU notebooks with an OpenAI-compatible local client. Spins up a private Kaggle kernel, installs Ollama, opens a Cloudflare tunnel, and exposes the running model through a standard API — no paid GPU required. Supports Strands Agents and LangChain integrations out of the box.

Python Ollama Kaggle GPUs Strands LangChain

003

AI Expense Management Agent (In Progress)

Developing an agentic system that autonomously analyzes receipt data, tracks spending patterns, and assists with budgeting decisions. The agent performs multi-step reasoning using tools such as transaction databases and financial rules to generate insights, detect anomalies, and provide proactive recommendations.

Strands LLM agents Amazon Bedrock Guardrails
Coming soon

Expertise & tools.

Core areas

——

Deep Learning

MLOps

NLP & LLMs

Computer Vision

Distributed Training

Python

Advanced

PyTorch

Advanced

Hugging Face Transformers

Advanced

Computer Vision (object detection, OCR pipelines)

Proficient

Retrieval-Augmented Generation (RAG)

Proficient

LLM application development / LLM agents

Proficient

MLflow

Proficient

Vector search (e.g., FAISS)

Proficient

LangChain / LLM orchestration

Proficient

Let's build something worth knowing.

Open to AI/ML roles, research collaborations, and technical consulting. I respond within 24 hours.