Work

Ten years, in order

Machine learning and NLP since 2016 — Arabic speech first, then summarization and knowledge graphs, now agentic LLM systems. Where a role produced a paper, the line beneath it says so.

Email me for a CV

Dec 2025 — Present

Founding AI Engineer, Stealth Startup

  • Built and deployed LLM systems for 14+ isolated enterprise customers.
  • Co-developed a LangChain Deep Agents multi-agent system with memory management for structured and unstructured data retrieval, SaaS integrations, and report generation.
  • Automated enterprise data onboarding for AI use through schema discovery, embeddings, and metadata preparation, with fault tolerance — reducing onboarding time from hours to minutes.
  • Enhanced document generation and editing with reusable skills, validation, and self-repair loops.
  • Improved retrieval quality using hybrid search, file system facets, and re-ranking.
  • Built a Composio-based SaaS integration for governed access to Google Workspace, Slack, Notion, GitHub, and Linear, with RBAC, human-approval gates, and white-labelled OAuth.
  • LangChain
  • LangGraph
  • Claude Agent SDK
  • RAG
  • Composio

Nov 2020 — Dec 2025

Senior Data Scientist (NLP), Implicit(formerly Agolo)

  • Built Graph RAG and question-answering functionality for Implicit’s Answers product.
  • Introduced LLMs for summarization, knowledge graph construction, and question answering.
  • Published FactAlign, a knowledge-graph-based hallucination detection system.
  • Fine-tuned LLMs on KG-to-text for entity-centric summarization, surpassing the state of the art. Read the write-up ↗
  • Maintained and expanded the extractive summarization platform for wider use cases.
  • Knowledge Graphs
  • Neo4j
  • Graph RAG
  • PyTorch
  • HuggingFace

Mar 2019 — Oct 2020

Data Scientist, DataPlus

  • Developed and deployed dashboards and ETL/predictive-analytics workflows for financial, telecom, and government clients.
  • Tableau
  • Alteryx

Apr 2018 — Aug 2018

Google Summer of Code Student, Distributed Red Hen Lab

  • Implemented an Arabic speech recognition system for broadcast speech using Kaldi and VariKN.
  • Implemented an Arabic dialect identification system using Kaldi, Scikit-learn, and Keras.
  • Mentored several projects in subsequent GSoC programs.
  • Kaldi
  • VariKN
  • Scikit-learn
  • Keras

Nov 2017 — Nov 2018

Research Assistant, Cairo University, Faculty of Engineering

  • Co-developed Senteech, a speech emotion classification system for customer service calls in low-resource languages.
  • Scikit-learn
  • TensorFlow
  • openSMILE

Feb 2017 — Apr 2017

Teaching Assistant, Cairo University, Faculty of Engineering

  • Taught CMP302 — Advanced Algorithms.

Sep 2016 — Aug 2017

Machine Learning Researcher, RDI Egypt

  • Enhanced mispronunciation detection accuracy for Hafss, a Holy Quran recitation training system, by training new Kaldi models based on state-of-the-art architectures and incorporating new data. Hafss ↗
  • Maintained Hafss’s C++ language model toolkit, fixing and expanding its generation rules.
  • Developed a .NET data annotation tool for the linguistics team.
  • Kaldi
  • C++
  • .NET

Toolkit

Programming
  • Python
  • Java
  • SQL
  • Bash
ML / NLP
  • PyTorch
  • HuggingFace
  • Scikit-learn
  • LangChain
  • LangGraph
  • Claude Agent SDK
  • RAG
  • Knowledge Graphs (Neo4j)
Web and Cloud
  • FastAPI
  • Litestar
  • Flask
  • Google Cloud (App Engine, Cloud Run, Cloud Tasks)
  • Firebase/Firestore
Languages
  • Arabic (native)
  • English (proficient)