Reema Rafeek — AI Engineer, ML Engineer & Software Engineer
AI Engineer.Software Engineer.
18 projects · most recent first
Selected work
Full index
View all projects →Experience
Experience
Jun 2026 — Present · Full-time
AI Specialist
Level Up Training Center
- Automated NCLDP daily reporting, reducing preparation time from 2–3 hours to 10–15 minutes.
- Built the LMS portal and follow-up system with role-based access, WhatsApp follow-ups, payments and digital assessments.
- Deployed a live AI chatbot for customer enquiries; [REPLACE: conversations per month].
Aug 2025 — Present · Freelance
Freelance Web Developer
RR Media
- Delivered rrmedia.ae, including animated intro, video hero, service pages and an enquiry chatbot.
- Delivered client websites including Champion Sports Club across three branches.
Feb — May 2026 · Internship · Remote
AI Android App Intern
Mind Matrix
- Built the Road Health Tracker: a model classifies road condition, with offline storage in Room.
- [REPLACE: accuracy, users or test drives]
Jan — Apr 2026 · Internship
iOS Development Intern
Abu Dhabi Islamic Bank (ADIB)
- Built iOS screens in Swift and Xcode for core application flows.
- Delivered work in two-week sprints with user stories and task boards in a regulated environment.
Aug — Oct 2025 · Research internship
Machine Learning Intern
NIT Calicut
- 92% prediction accuracy on held-out test data.
Feb — Mar 2023 · Internship
Data Science Intern
Academor × Flutura
- Internship completion certificate, ID ACM23-557.
Interactive · optimization lab
Gradient descent
Gradient descent: an interactive optimization labHow every neural network learns. Drop a ball on the loss surface, pick an optimizer and tune the learning rate — or race all four optimizers from the same start.
LV 1 · Convex bowl
A convex surface: any starting point converges. Raise the learning rate to see overshoot.
v ← 0.9v − η·∇L ; θ ← θ + v
- The gradient
- The direction in which the error rises fastest. Training steps the opposite way, downhill, a little at a time — the arrows in the ∇ field.
- Learning rate η
- How far each step goes. Too small and training crawls; too large and it overshoots and diverges. Choosing it well saves hours of compute.
- Optimizers
- Momentum carries speed through flat regions; RMSProp and Adam scale each direction separately, which is why they handle narrow ravines and are the default for deep networks.
Technical stack
All skills →About
AI engineer building production LLM and machine learning systems.
I design and build AI systems that are used in production: a bilingual RAG assistant serving customers across four channels, a learning portal and follow-up system for a training centre, and the web platforms that deliver them.
I work end to end, from data and model evaluation to APIs, interfaces and automated testing, with a focus on reliable, measurable outcomes.
Contact