Omar Elsaber

AI Engineer specializing in Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and applied Machine Learning, with a proven record of architecting and shipping production-grade AI systems end-to-end — from data pipelines to LLM orchestration and deployment. Solo founder building AI-powered B2B SaaS products, pairing RAG and agentic AI architecture with strong MLOps fundamentals (FastAPI, Docker, CI/CD, MLflow) to deliver measurable outcomes: 90%+ model accuracy, sub-20ms latency, and hallucination-resistant LLM pipelines validated against strict quality gates. Award-winning researcher (RoboCup Asia-Pacific 2024 Bronze Medalist) delivering scalable, high-accuracy AI solutions under real-world constraints.

Egyptian Legal AI Assistant (Arabic RAG System) at Independent Project (2026-01 – 2026-04)

• Architected a production-grade Arabic RAG pipeline using LlamaIndex + ChromaDB + Groq (Llama3-70B), implementing a 5-stage query pipeline — colloquial-to-formal query rewrite, top-20 cosine retrieval (Cohere multilingual embeddings), BAAI cross-encoder reranking, citation formatting, and strict Arabic synthesis — delivering grounded, hallucination-resistant legal answers.
• Engineered an end-to-end LLMOps evaluation framework using Ragas + MLflow, enforcing quality gates (Faithfulness ≥ 0.7, Answer Relevancy ≥ 0.6) as CI/CD blockers via GitHub Actions — automatically blocking merges on metric regression and logging per-request latency, token counts, and eval scores to MLflow.
• Designed a multilingual Arabic-first embedding pipeline with asymmetric Prompt Engineering (separate search_document / search_query input types), improving cosine retrieval precision by ~15–20%; served via FastAPI with Pydantic-validated schemas, a React + Tailwind RTL-native frontend, and Docker-based infrastructure for ChromaDB and MLflow.

Machine Learning Researcher – EEG Signal Classification (BCI Challenge) at Kaggle

• Designed end-to-end signal processing pipelines for Brain-Computer Interface classification — bandpass filtering, normalization, and feature extraction at scale across multi-channel EEG datasets; benchmarked ensemble strategies (CCA + eCCA for SSVEP, LightGBM for Motor Imagery), achieving 79% accuracy, surpassing Kaggle competition baselines in EEG signal decoding.

End-to-End Used Car Price Prediction System at Independent Project (2025-11 – 2026-01)

• Engineered an end-to-end MLOps pipeline using XGBoost (1,000 trees) with MLflow experiment tracking, achieving 93% R² accuracy (vs. 29% baseline) — a 132% performance improvement driven by data quality migration (Cardekho dataset) and feature engineering (brand extraction, car age, mileage-per-year normalization).
• Built a production microservices architecture with FastAPI backend + Streamlit frontend, fully containerized via Docker Compose with sub-20ms prediction latency; automated CI/CD via GitHub Actions covering linting (Flake8/Black), type checking, and 38+ Pytest unit tests at 92% code coverage.
• Implemented full data versioning with DVC, scikit-learn ColumnTransformer preprocessing pipeline, L1/L2 regularization, and log-target transformation for skewed distributions; validated all API inputs with Pydantic schemas for production-grade data integrity.

Predicto — AI-Powered Revenue Intelligence Platform at Independent Project (2025 – Present)

• Architected and solo-built Predicto, a B2B SaaS revenue intelligence platform (FastAPI + React 19/TypeScript) that ingests CRM/RevOps data, trains ML models at ingest time, and serves forecasts, deal scoring, and churn signals via an executive dashboard.
• Engineered three production ML pipelines — Fourier+Ridge forecasting, XGBoost deal scoring, K-Means segmentation — plus a hybrid PyTorch GRU + XGBoost cold-start router for sparse data, exposed across 30+ REST endpoints covering 7 RevOps KPIs and DML-based causal analysis.
• Integrated Groq (Llama 3.3-70B) for SSE-streamed executive narratives via a token-budgeted ML-to-LLM context builder; shipped a bilingual (EN/AR RTL) dashboard with TanStack Query and Tremor/Recharts.

B.Sc. in Computer Science — Artificial Intelligence Major – Arab Open University (2021-09 – 2025-05)