Available for new projects

AI Engineerbuilding thingsthat learn.

I build and ship AI systems that run in production: RAG systems, fine-tuned Arabic LLMs, and multi-modal edge inference for human–robot interaction.

View projectsGet in touch

1.5s

RAG time-to-first-token

443

tok/s via vLLM

97%

face-recognition accuracy

70%

workflow time cut

Selected work

Projects

01
1.5s TTFT

Internal RAG Chatbot

Internal knowledge chatbot on AWS ECS Fargate with an EFS-backed vector store. 1.5s TTFT, 99.9% uptime, plus dashboards tracking query latency and token usage.

ChromaDBLangChainFastAPIAWS ECS
02
443 tok/s

Arabic News LLM Fine-Tuning

Fine-tuned Qwen2-1.5B with LoRA for Arabic news summarization and entity extraction, then served it with vLLM at 443 tokens/second.

Qwen2LoRAvLLMArabic NLP
03
7s latency

Human–Robot Interaction (Edge)

Ran SmolVLM2 video inference on Jetson Nano / Raspberry Pi at 7s end-to-end latency for a live VLM + TTS + ASR + LLM system.

SmolVLM2Jetson NanoEdge AIMulti-modal
04
97% accuracy

AI People Tracking

Led a 9-person team on a mall surveillance system; built the Siamese-network face-recognition module at 97% accuracy to help locate lost children.

Siamese NetFace RecognitionTeam Lead
05

Arabic Sign Language Classifier

Transfer-learning classifier (ResNet50V2) for the Arabic sign-language alphabet, with a data-augmentation pipeline to handle class imbalance on limited data.

ResNet50V2Transfer LearningAccessibility
06

TimeMe Chrome Extension

Per-tab time-tracking Chrome extension with productivity analytics. A shipped end-user product, built front to back.

JavaScriptTailwindProduct

Where I've worked

Experience

AI Engineer & Instructor

Saudi Specialized Training & Learning Institute (SSTLI)

Mar 2025 — Present

Dammam, Saudi Arabia

  • Shipped an internal RAG chatbot (ChromaDB + LangChain + FastAPI) on AWS ECS Fargate — 1.5s TTFT, 99.9% uptime.
  • Built dashboards tracking query latency, token usage, and sales KPIs in real time.
  • Optimized SmolVLM2 edge inference on Jetson Nano / Raspberry Pi to 7s for a live human–robot system.
  • Automated manual workflows with Selenium, cutting task time by 70%.
  • Taught AI courses to diploma students and business professionals — explaining transformers and vector DBs in plain language.

Freelance Machine Learning Engineer

Upwork

Jul 2024 — Present

Remote

  • Deliver end-to-end ML systems for international clients across NLP, computer vision, and geospatial data science — scoping, building, deploying, and owning them in production.
  • Owned delivery of a geospatial business-prediction platform for the Saudi market: ingestion from the Google Places API and OpenStreetMap, XGBoost tracked with MLflow, backend APIs, and deployment.
  • Built the feature pipeline on PostgreSQL + PostGIS as the single source of truth for training and serving — 35K feature rows in ~25s via set-based SQL.
  • Deployed the production stack on AWS EC2 — Docker Compose running FastAPI, PostgreSQL 16 + PostGIS, and Redis on one cost-optimized instance.
  • Built a GitHub Actions CI/CD pipeline that lints, tests against an ephemeral Postgres service, and publishes a multi-stage image to GHCR.

ML Engineer Intern

Digital Egypt Pioneers Initiative (DEPI) — Microsoft Partner

Oct 2024 — Mar 2025

Cairo, Egypt

  • Built an AI-powered predictive-maintenance pipeline on the NASA bearing dataset, tracked with MLflow and deployed on Azure.
  • Completed a 6-month program in ML/DL, NLP, CV, Transformers, Generative AI, and Azure.

Toolkit

Skills & Stack

AI / LLMs

  • PyTorch
  • Transformers
  • LoRA Fine-tuning
  • RAG
  • LangChain
  • AI Agents

Product Eng

  • Python
  • FastAPI
  • Flask
  • Pydantic
  • REST APIs
  • n8n

MLOps & Infra

  • Docker
  • CI/CD
  • MLflow
  • vLLM
  • llama.cpp
  • Triton
  • CUDA

Cloud & Edge

  • AWS (ECS/EFS/Bedrock)
  • Azure
  • GCP
  • Jetson Nano
  • Raspberry Pi

Background

About me

I'm an AI Engineer focused on systems that ship and run in production — RAG systems, AI agents, fine-tuned Arabic LLMs, and multi-modal edge inference.

Teaching AI courses keeps me in daily contact with students, instructors, and admins, so I'm used to explaining transformers, vector databases, and quantization to non-technical people quickly.

I work with Claude Code and Cursor to move from problem to spec to working prototype fast. Fluent in Arabic and English (C2), AWS AI Practitioner certified.

B.Sc. Electronics & Communications Eng.Valid SCE Registration (IT/Telecommunications)AWS Certified AI PractitionerML Specialization — Stanford / DeepLearning.AICS50x — Harvard

Currently

Based in Dammam, Saudi Arabia

Find me

Let's talk

Let's build something
intelligent.

Have a model to train, a pipeline to ship, or an idea worth prototyping? I'd love to hear about it.

ibrahem.elsherif@ibrahemelsherif.com