Fahad Dubush

AI Engineer. I explore how AI and automation can turn unconventional ideas into practical solutions that save time, reduce costs, and make technology work smarter.

Curriculum Vitae CV

Selected work

About me

I’m drawn to ideas that sit outside the usual playbook: things that haven’t been tried yet, or opportunities others might overlook. I enjoy exploring how AI and automation can solve problems in ways that save time and money. Before starting a project, I like to research the options and find the best balance between capability and cost. My curiosity extends beyond technology to games, films, shows, and books, anywhere a new idea or perspective can spark something worth building.

Let’s talk.

Case studies

Pilgrim Companion

AI guidance for Hajj and Umrah. A local-first question-answering application that retrieves multilingual sources and grounds responses in cited context.

Pilgrim Companion interface showing the question prompt and grounding panel.
Pilgrim Companion interface

Retrieval pipeline

A simplified view of the pipeline described in the project README.

  1. English or Arabic question
  2. Domain routing
  3. Dense and BM25 retrieval
  4. Rank fusion and relevance checks
  5. Optional reranking
  6. Grounded prompt and local or API model
  7. Streamed response with sources

Retrieval evaluation

Results reported in the repository for a 12-question English and Arabic regression set on a CPU development machine. These results are not a measure of religious correctness or production accuracy.

MetricResult
Retrieval Recall@4100%
Mean Reciprocal Rank0.926
Out-of-domain rejection100%
Median retrieval latency51.8 ms
Knowledge-base chunks123 unique chunks
  • Combines semantic and keyword retrieval for English and Arabic questions.
  • Supports local Ollama models and OpenAI-compatible APIs.
  • Includes source citations, streaming responses, and out-of-domain handling.
  • Uses conversation-safe caching for standalone questions to avoid reusing cached results for follow-up questions.
  • Python
  • FastAPI
  • RAG

View source

Back to work

Groundtruth

A retrieval evaluation toolkit for legal research. An evaluation harness for comparing retrieval approaches and detecting regressions in a legal research dataset.

Diagram showing lexical and dense retrieval, rank fusion, reranking, and a CI evaluation gate.
Groundtruth retrieval architecture

Retrieval benchmark

Results reported in the project README for its 100-query evaluation set. Benchmark results are not evidence of legal safety or production readiness.

ConfigurationRecall@5Recall@10MRR@10nDCG@10Mean latencyp95 latencyHard failures
BM250.7880.8480.9640.9130.3 ms0.4 ms1 / 100
Dense0.8280.9070.9730.9265.3 ms6.4 ms0 / 100
Hybrid0.8380.8820.9880.9395.8 ms6.9 ms0 / 100
Hybrid + reranker0.8620.8980.9810.93922.2 ms27.7 ms0 / 100
Hybrid, fine chunks0.8320.8880.9750.9245.8 ms7.0 ms0 / 100
  • Compares lexical, dense, hybrid, and reranked retrieval approaches.
  • Reports retrieval metrics and results by practice area.
  • Includes a CI gate designed to flag retrieval regressions.
  • Python
  • RAG
  • CI

View source

Back to work

FORGE Athletic Apparel

A deployed website for an athletic apparel client. A client project for FORGE Athletic Apparel, built as a deployed coming-soon website for the brand.

FORGE Athletic Apparel logo on a dark background.
FORGE brand preview
  • Created for an athletic apparel client.
  • Deployed at forgeksa.shop.
  • Presents the brand through a dedicated coming-soon web page.
  • HTML
  • Web Design

Visit websiteView source

Back to work

HomeLab TUI

A terminal dashboard for a self-hosted server. An SSH-friendly Linux terminal interface for monitoring host telemetry, Docker containers, and maintenance tasks.

HomeLab terminal dashboard in demo mode with fictional telemetry.
HomeLab demo with fictional data

Dashboard overview

Features described in the repository README. Availability depends on host configuration and permissions.

AreaWhat it shows or does
OverviewCPU, memory, network, disk usage, and container status
ContainersDocker container states and selected container actions
UpdatesCompose checks and eligible update options
Storage, network, securityHost diagnostics and checks
Backups, logs, maintenanceExplicitly invoked tasks, logging, and reporting
SettingsAnimation, polling, update pacing, and help
  • Brings host metrics, container status, logs, and maintenance tools into one interface.
  • Includes a demo mode with fictional data and disabled actions.
  • Supports configurable settings, keyboard navigation, and optional animation.
  • Go
  • Docker
  • Linux

View source

Back to work