متدرب في تعلم الآلة (معالجة اللغات الطبيعية باللغة العربية)

Company Description:

Finalyze is a Cairo-based data company building clean financial data for emerging markets, starting with the Egyptian Exchange. Much of the market's data still sits in Arabic filings that analysts retype by hand. We extract the financial statements, flag figures that don't add up, and deliver normalised data in English that can be traced back to the exact page it came from. We are a small founding team and a graduate of the American University in Cairo's Venture Lab.

Role Description:

We are hiring one machine learning intern for a 12-week, part-time, remote internship on our capital markets data product. You will work on two things: measuring how accurately the large language models behind our extraction read Arabic financial statements, and building a sentiment score for each company listed on the Egyptian Exchange. You will write a one-page plan before you start, every result you report has to be rerunnable by someone else from your notebook, and every change is reviewed by our chief technology officer before it merges.

What you will work on:

• A test of how reliably the models behind our extraction read Eastern Arabic numerals

• A comparison of those models' accuracy against our bar, and of their cost per statement

• An evaluation set built from verified pages, so every change to the pipeline is measured before it ships

• A sentiment score for each listed company, built from text ingestion, a labelled dataset and a baseline model

• Stack: Python and notebooks, with sentiment built as a new service.

Terms:

• Starts November 2026

• 12 weeks, 15 to 20 hours a week, including two fixed afternoons each week for review with the team

• Fully remote

• This is an unpaid internship

Requirements

• Third year or above in computer science, data science or a related field, a master's student, or a recent graduate

• Reads Arabic fluently

• Experience training a text classifier

• Experience measuring a model's output against labelled data, for example with precision and recall

• Comfortable in Python and notebooks

• Able to commit 15 to 20 hours a week for 12 weeks, including two fixed afternoons each week

• Comfortable working remotely and writing down your plan before you build