┌─ majid.png @ 32×24 · amber phosphor ─┐
MAJID
KOFIA
ORIGIN
A kid in Jeddah who got addicted to making machines think.
▌ JEDDAH · 21.49°N
Curiosity arrived early. The kind that breaks things to see how they work — old PCs, broken routers, a Game Boy that never came back together. Then code showed up, and finally an ML lecture in 2021 that rewired everything.
Now I build systems that take messy inputs — pdfs, photos, voice notes, financial feeds — and turn them into something usable. LLMs, RAG, vision, automation. The plumbing of intelligence.
Speaks Arabic (native) & English (IELTS 6.5 / B2). Drinks too much karak chai.
THE ACADEMY
Five years at the University of Jeddah, decoding intelligence.
Hardware tinkering, flight controllers, autonomous nav experiments. Learned to crash gracefully — and rebuild faster.
Weekly study sessions, paper readings, hackathons. The room where most of my real learning happened.
TRIALS
Two internships in 2024 — testing what I'd learned in the wild.
- ▸Building production AI systems for private clients — clinical automation, financial intelligence, and document pipelines.
- ▸End-to-end ownership: architecture, RAG, model integration, infrastructure, deployment, and handover.
- ▸Automated resume parsing: built a PDF → structured CSV pipeline, killing hours of manual entry.
- ▸Engineered prompts for an LLM-powered HR system; improved candidate-screening accuracy.
- ▸Trained an unsupervised topic model on employee feedback to surface themes ops cared about.
- ▸Optimized batch workflows with GPU parallelization for faster model inference.
- ▸Worked tickets across departments, owning escalation workflows end-to-end.
- ▸Configured and imaged PCs with secure enterprise setups; debugged NOC-switch infrastructure.
QUESTS COMPLETED
Things I've shipped. Some live in production. All were built end-to-end.
ClinicalBot
WhatsApp AI assistant
Production WhatsApp chatbot for clinical-report management. Gemini 2.5 Flash with function calling — zero hard-coded intent detection.
RAG pipeline with pgvector + Supabase for semantic search across medical document chunks. Multi-node n8n orchestration. Dockerized microservices.
Market Intelligence Newsletter
Automated weekly financial brief
Scrapes Saudi financial news, classifies articles by portfolio relevance with Gemini, and ships analyst-grade equity summaries.
Seasonal context awareness — Ramadan, Hajj, earnings windows — to surface indirect market impacts. Premium HTML email with live TASI + Brent data.
Clinical Report Reconstruction
Lab & imaging report rebuild pipeline
Internal Next.js webapp for a private clinic's staff. Extracts structured data from incoming lab and imaging PDFs heuristically with pdfjs-dist, lets staff review and edit the parse, then re-renders the full report from scratch — categorized rows, abnormal-value flags, bilingual Arabic/Latin typography, and clinic branding baked in.
Replaces an earlier overlay-based rebrander with a proper data → PDF reconstruction pipeline. pdf-lib for layout, fontkit for Arabic shaping, all server-side in Next.js route handlers.
Saudi Investment Dashboard
Private portfolio monitor
Real-time dashboard for a private Saudi equity portfolio — NADEC, Al Rajhi Bank, Yanbu Cement, TASI, and Brent crude. Tracks quotes, disclosures, news, and oil benchmarks in one view.
FastAPI backend + React/Vite frontend. Aggregates yfinance, Twelve Data, Tadawul scrapers, RSS (Argaam, Arab News, SPA, OPEC), and EIA oil data. Claude generates weekly summaries with portfolio context.
Football Analytics Platform
Saudi AI League
End-to-end football analytics: YOLOv8 for real-time player/ball detection and tracking + LLaMA 3 for tactical insights and match reports. Streamlit front.
Player Recommendation System
Senior Project
Recommender on FIFA player attributes (rating, position, weak foot) using cosine similarity. Optimized preprocessing pipeline.
ARSENAL
The kit I bring to every quest.
RECRUIT
Looking for the next quest. Hiring? Got a weird idea? Send word.