
Pearls Consulting
TunisiePublication : 19 nov. 2026
INTERNSHIP: AI SOFTWARE ENGINEER
- Stage
- 4-6 mois
- Hybride
- Rémunéré
- Échéance : 3 déc. 2026
- Artificial intelligence and data science
- Software engineering
- Machine learning
- AI/LLMs
- Computer vision / 3D scanning
- NLP/Document AI
- OCR
- Excel/Python
- Backend (FastAPI)
- Frontend (Angular/React)
- Python/TypeScript
Description du poste
Internship
Pearls Consulting is offering a paid PFE internship as an AI Software Engineer. The internship is based in Tunisia, at the company's Lac 2 location, and involves building AI tools for understanding real business documents.
Project context
As part of the engineering team, you will help build AI tools that read and understand:
- Commercial registrations
- Financial statements
- Certifications
- Contracts
- Policy frameworks
The work combines frontier LLM APIs with open-weight models that are fine-tuned and deployed in-house. The tools cover one or more of the following functions:
- Procurement and Vendor Management
- Finance and Accounting
- Regulatory Compliance
- Operations
Main responsibilities
- Experiment with open-weight models such as Llama, Qwen, and Mistral, and benchmark them against frontier LLMs including Anthropic Claude and OpenAI GPT.
- Fine-tune vision-language models for document understanding, including scanned pages, tables, forms, and stamps.
- Build evaluation datasets and pipelines to measure accuracy before and after fine-tuning.
- Design prompts and structured extraction for LLM-based document processing.
- Help deploy and serve models, and integrate them into PDF, Excel, and OCR pipelines.
- Build features across the stack using Python, FastAPI, React, and TypeScript to bring the AI tools to users.
Requirements
- Currently pursuing a degree in Computer Science, Software Engineering, AI, or Data Science, with a strong academic record.
- Good grasp of machine learning fundamentals, including training, fine-tuning, and evaluation metrics.
- Solid Python skills and familiarity with PyTorch and the Hugging Face ecosystem, including Transformers and PEFT.
- Strong problem-solving skills, curiosity about how models work under the hood, and a willingness to learn.
- Experience fine-tuning LLMs or vision models using LoRA or QLoRA, OCR/document AI, or working in GPU and Linux environments is a plus.
Languages
English and Arabic are mandatory.