Ubeid Teelhawod

I’m a final-year Data Science student in Mauritius, working towards a career as a data and AI engineer.

I design data models and the SQL databases behind them, and use them to answer real questions.

About

I’m studying for a dual degree in Data Science, at the University of Mauritius and CY Cergy Paris Université.

Alongside it I study accounting with ACCA, which is why I’m drawn to problems where data meets finance. My final-year project, a tool that assesses small-business loans fairly, is one of them. My work centres on Python, SQL and database design: data pipelines, back-end systems and applied machine learning. I want to build the data infrastructure that lets organisations use AI well.

Studying: Dual degree, 2024–2027
Also: ACCA, in progress
Based in: Phoenix, Mauritius
Open to: Part-time or flexible work

Skills

Each tool with the work where I used it.

Languages

PythonBDO, MNIST, SME agent
SQLBDO, ElectroShop, AEGIS
RStatistics coursework
JavaScript, PHPPaws and Claws
Java, CCoursework

Databases

SQL ServerBDO
MySQLElectroShop, SME agent
SQLiteAEGIS
SQLAlchemyBDO

AI and frameworks

TensorFlow (Keras)MNIST
FastAPI, ReactSME agent
Ollama, GPT, GeminiSME agent
Tesseract OCRSME agent

Design and analytics

Power BI, TableauCertificate, crime dashboard
BPMN 2.0, BizagiBDO, coursework
FigmaBDO, Paws and Claws

Projects

Selected internship, coursework and hackathon work from 2024 to 2026.

Fair SME loan assessment agent

Final-year project, in progress

A lending app that helps assess small-business loans in Mauritius fairly, and explains each decision it suggests.

Built with FastAPI, React, MySQL, PyMuPDF, Tesseract, Ollama, GPT, Gemini

Built so far

SME profiles and loan application workflows
Role-based access control
OCR extraction from PDF documents with PyMuPDF and Tesseract
A chatbot that runs on Llama 3.2 locally through Ollama, or calls GPT and Gemini

ElectroShop data warehouse

Data Warehousing module, solo

A sales data warehouse in MySQL 8 for a fictional Mauritian electronics retailer: 218 transactions worth Rs 8,670,780.

Key highlights

Cleaned 357 source rows down to 351 with ETL rules written in SQL
Designed a star schema: one sales fact table and four dimensions
Answered five management questions in SQL, one query each
Ran the OLAP operations (roll-up, drill-down, slice, dice, pivot) and built CUBE-style roll-ups from UNION ALL, since MySQL has no CUBE operator
Found that television sales fell 27.5% from 2024 to 2025

Built with MySQL 8, SQL

Explore the sales

Sales by year, from the warehouse.
Year	Sales
2024	Rs 4,041,690
2025	Rs 4,629,090
Total	Rs 8,670,780

Star schema: the FACT_SALES table in the centre, joined by time_key, product_key, store_key and customer_key to DIM_TIME, DIM_PRODUCT, DIM_STORE and DIM_CUSTOMER. — The star schema. Open it full size.

Warehouse architecture: stores, sales, products and customers feed a load manager, which loads detailed sales data into the ElectroShop warehouse. A warehouse manager keeps monthly and yearly summaries and the metadata, and a query manager serves SQL queries, reports and OLAP tools. — The warehouse architecture, from four source files to the query tools. Open it full size.

AEGIS anomaly review database

Game of Code 2026, team of four, 12 hours

A prototype that flags unusual movement in public-space video and sends it to a human operator, who reviews it before anything happens. I owned the database behind that review workflow.

Key highlights

Designed a 15-table SQLite schema, with seed scripts and documentation
Linked every alert to its camera, anomaly type, AI score, severity, evidence, human review and final outcome
Kept a full review trail, so every decision can be traced

Built with SQLite. Teammates built the FastAPI back end, the Next.js interface and the PyTorch autoencoder.

Entity relationship diagram of the AEGIS database: 15 tables including locations, cameras, video_sources, anomaly_events, evidence_clips, anomaly_reviews, operators and ngos, joined by foreign keys. — The AEGIS database schema. Open it full size.

Handwritten digit classifier

Coursework

A neural network that reads handwritten digits with 97.58% accuracy on 10,000 test images it had never seen.

Key highlights

Two hidden layers of 128 and 64 units, trained for 10 epochs
Read the training curves to find where to stop: validation loss levels off after the fourth epoch
Studied the 242 digits it got wrong, eight of them shown here

Built with Python, TensorFlow (Keras)

Two line charts over 10 epochs. Loss: training falls from 0.348 to 0.020 while validation levels off near 0.08 after epoch 4. Accuracy: training rises from 90.2% to 99.4% while validation holds near 97.7%. — Loss and accuracy over 10 epochs, and eight digits the network misread.

Paws and Claws booking site

Web Programming, team of three

A booking site for a pet hotel. I built the booking flow and the accessibility features.

Key highlights

A booking form that validates every field, works out the price and saves each booking to CSV
Mobile navigation and responsive layouts
A high-contrast theme, with the wireframe checked against WCAG 2.2

Built with JavaScript, PHP, Figma

The finished booking page. Open it full size.

Experience

Two internships in the AI & Innovation team at BDO IT Consulting, Port Louis.

AI & Innovation Intern

May 2026 – Jul 2026

BDO IT Consulting

Redesigned the SQL Server database of an internal recruitment platform, extending it from about 28 tables to 55 to add role-based access control, teams and workspaces
Introduced BPMN 2.0 to the team and mapped HR and recruitment processes, so business and technical staff could plan automation from the same diagrams
Contributed to the wireframes for a digital lottery campaign, drafting with AI and refining the user journeys in Figma, then presented the method at the quarterly company meeting

Used SQL Server, Python, SQLAlchemy, BPMN 2.0, Bizagi, Figma

AI & Innovation Intern

Oct 2025 – Jan 2026

BDO IT Consulting

Built the back-end database of an internal AI-driven HR automation system, with migrations and seed data
Added fallback logic and audit logging, and tested each stored procedure three ways: success, fallback and complete failure
Contributed to the launch of an AI Readiness Assessment Tool that measures how ready an organisation is to adopt AI

Used Python, SQLAlchemy, SQL Server

Recommendation

“I had the opportunity to mentor Ubeid during his internship, and he consistently demonstrated professionalism and discipline, alongside a strong commitment to learning. … His reliability and willingness to take ownership of assigned tasks reflected a level of professionalism beyond that typically expected at an intern level.”

Mustaq Bengah

Senior Software Engineer (Data), BDO IT Consulting. My mentor at BDO.

Read the full recommendation on LinkedIn →

Education

University of Mauritius

2024 – 2027

BSc (Hons) Data Science, final year

CY Cergy Paris Université

2024 – 2027

Licence Informatique (Data Science), the second half of the dual degree

ACCA

Since 2024

Applied Knowledge level complete. Financial Management and Performance Management passed at Applied Skills.

Sir Abdool Raman Osman State College

Until 2023

Cambridge Higher School Certificate

Main subjectsA in Accounting, Economics and Mathematics

Certifications

Google AI Professional Certificate
Google, 2026
Data Analyst in Power BI
DataCamp, 2026
Building Your First AI Agent with OpenAI
Coursera, 2026
BPMN 2.0 with Bizagi
Udemy, 2026
Figma UI UX Design Essentials Pt. 1: Tools & Foundations
2026
Claude Code 101
Anthropic, 2026
Claude 101
Anthropic, 2026
Claude Platform 101
Anthropic, 2026
Introduction to Claude Cowork
Anthropic, 2026

Contact

I’m looking for part-time or flexible work in AI and data.

teelhawodu857@gmail.com Open in Gmail

GitHub LinkedIn CV (PDF)