Yousef Alsihli

AI Engineer · Riyadh, Saudi Arabia

I build agentic systems: LLM orchestration, evaluation harnesses, and AI products that have to survive contact with real operations.

Most of my work sits in the layer around the model. Memory, retrieval, routing, deterministic tools, verification. That layer is usually where the gap between a demo and something dependable actually lives.

I care about proving things. If I claim a system got better, there is a reproducible evaluation behind the number. I also build Arabic first, with real RTL, because most tools still treat it as an afterthought.

Projects

  1. Powerhouse AI

    2026

    Private local LLM orchestration and evaluation.

    An open-source orchestration system for Windows and NVIDIA hardware that wraps one private local model in project memory, document retrieval, specialist routing, deterministic tools, synthesis and final verification.

    I built a reproducible 44-case evaluation harness to measure it honestly. On the same Qwen3.5-9B (Q6_K), results moved from a 76% baseline to 94% in Fast mode and 97% in Adaptive mode. Released MIT with source, automated tests, benchmark evidence and validation docs.

    76% → 97% on a 44-case evaluation, same model

    • Python
    • FastAPI
    • RAG
    • LLM Evaluation
    • AI Agents
    github.com/kkmmee94/powerhouse-ai ↗

  2. Renavo

    2026

    AI operations layer for hotels in Saudi Arabia and the GCC.

    Founded and built solo. It turns guest requests into traceable tasks with guarded actions, approval holds, human escalation and auditable outcomes, bilingual in Arabic and English with complete RTL throughout.

    Under it sits secure guest sessions, idempotent writes, verified readback, an operational ledger and a persistent Test PMS for end-to-end evaluation. The public demo runs on synthetic data so the whole workflow can be inspected safely.

    Live product, publicly testable on synthetic data

    • Next.js
    • TypeScript
    • Supabase
    • LLMs
    • AI Agents
    renavo.net ↗

  3. Arabic RTL Rendering Engine

    2026

    Arabic localisation for the game Oxygen Not Included.

    A C# Harmony runtime renderer handling connected Arabic shaping, RTL layout and mixed Arabic/Latin text across TextMeshPro, while protecting URLs, versions, units, scientific notation and editable fields from being reversed.

    Coverage, placeholder, rich-text, numeral and release-packaging checks all run in GitHub Actions.

    21,122 entries validated · 62 regression assertions

    • C#
    • Harmony
    • Software Localization
    • GitHub Actions
    github.com/kkmmee94/oxygen-not-included-arabic ↗

  4. Salama

    2026

    Arabic-first smart-home maintenance platform for iOS.

    Built in SwiftUI, SwiftData and Swift Charts with complete RTL localisation, offline operation, notifications, cost analytics and multi-home support, backed by a bilingual Express and SQLite service.

    28 unit tests across scoring, scheduling, localisation and API contracts

    • SwiftUI
    • SwiftData
    • Express
    • REST APIs

  5. UoM WAM Calculator

    2026

    Privacy-first academic planning, fully offline.

    Models subjects, semesters and credit points to project WAM, calculates the marks required to hit a grade target, and flags outcomes that are mathematically unreachable. Supports Course Planner imports, scenario modelling and local backups. Nothing leaves the device.

    91 automated tests behind the calculation engine

    • Data Science
    • Offline-first
    • Scenario modelling
    github.com/kkmmee94/uom-wam-calculator ↗

Experience

  1. Founder & AI Product Engineer, Renavo

    Jul 2026 – Present
    • Founded and independently built an AI operations layer for hotels in Saudi Arabia and the GCC.
    • Designed bilingual guest and operational experiences with complete RTL support and shared real-time state.
    • Built concierge workflows converting guest requests into traceable tasks, guarded actions, approval holds and human escalations.
    • Implemented secure sessions, idempotent writes, verified readback, an operational ledger and a persistent Test PMS.

  2. AI & Data Product Engineer, BIT TECH

    May 2024 – Present
    • Integrated a human-in-the-loop workflow that analyses patient-provided medical images into structured recommendations for physician review. It supports the clinician and never confirms an outcome without doctor approval.
    • Identified and helped remediate security vulnerabilities in Doctoria, a production telehealth platform on iOS and Android.
    • Profiled performance, analysed bottlenecks and shipped targeted optimisations, improving measured screen-loading performance by 2.5×.
    • Contributed to architecture decisions, development planning and release quality across the mobile and web portfolio.

Skills

AI and machine learning

  • Artificial Intelligence
  • Large Language Models
  • AI Agents
  • Retrieval-Augmented Generation
  • LLM Evaluation
  • Generative AI
  • Natural Language Processing
  • Machine Learning
  • Intelligent Automation

Languages and frameworks

  • Python
  • TypeScript
  • C#
  • SQL
  • FastAPI
  • Next.js
  • React.js
  • SwiftUI

Data, platform and craft

  • Data Science
  • Supabase
  • REST APIs
  • Git
  • Performance Tuning
  • Web Application Security
  • Software Localization
  • Software Development

Education

  1. BSc Data Science, University of Melbourne

    Jul 2025 – Jun 2028

    Major in Data Science, minor in Artificial Intelligence. Machine learning, statistical modelling, algorithms and data systems, applied through the products above.

  2. Foundation Studies (Science), Trinity College, University of Melbourne

    Jul 2024 – Jun 2025

  3. High School Diploma, Al-Shuhudaa

    Sep 2019 – Jun 2023

    Graduated with a grade of 99.03.

Recognition

  1. College Degree Program Scholar, Saudi Aramco

    Aug 2023

    Selected through an approximately 1% acceptance process. Completed the intensive College Preparatory Program in Dhahran, covering advanced mathematics, computing and university preparation, earning 5/5 in AP Computer Science A and 5/5 in AP Calculus AB.

  2. CS50’s Introduction to Computer Science, edX

    Jan 2024