AI Research Intern - Evaluating and Improving a RAG Pipeline for Heterogeneous Documents

The Internship In Short

Join the artificial intelligence team at MULTITEL. During this internship, you will work on Retrieval-Augmented Generation 
(RAG), a paradigm for deploying large language models on private, domain-specific knowledge. 
By grounding model answers in a knowledge base built from an organization’s internal documents, RAG makes it possible
to answer questions accurately while preserving data residency and confidentiality, which is a key requirement for 
on-premise deployments.

At MULTITEL, the RAG stack runs entirely on-premise. It is built around a chat interface that lets users create a knowledge
base by uploading documents and ask questions whose answers are grounded in the corpus. However, the stack 
currently lacks a robust evaluation pipeline. Without a way to measure retrieval and generation accuracy, it is hard to 
know whether a change improves the system, or to improve it further in a principled way.

The topic

You will play a key role in building an evaluation framework for the RAG pipeline and using it to improve the system across a wide range of document types. The work has two objectives: first, to make RAG performance measurable with open-source evaluation tools; second, to use those measurements to tune the pipeline so it handles each document type reliably.

The system must handle many document types, several particularly challenging for RAG:

• Two-column scientific papers with embedded images and graphs break reading-order detection and chunking, losing content.
• Scanned and handwritten documents need robust OCR, as poor recognition degrades the retrievable text.
• Long regulatory documents with many cross-references stress chunking, since answering often requires assembling information across the whole document.
• Documents with large tables challenge extraction and embedding, as tabular content rarely converts cleanly to flat text.

Improving the pipeline on each category is a central objective of the internship.

What you will work on:

• As an intern, you will play a key role in making RAG performance measurable and then improving it. Responsibilities include:
• Selecting and setting up an RAG evaluation framework using open-source tools (e.g. RAGAS, Deepeval), including the construction of representative evaluation datasets for each document category; 
• Defining and computing relevant metrics for both the retrieval and the generation stages (e.g. faithfulness, answer relevance, retrieval precision, context recall);
• Auditing the document extraction and preprocessing pipeline and tuning its parameters to improve the handling of challenging document layouts;
• Evaluating the embedding model, chunking strategy, and retrieval configuration, and proposing improvements backed by quantitative results;
• Benchmarking the pipeline before and after each change to demonstrate measurable improvement;
• Synthesizing key findings into a detailed report;
• Presenting results to the AI team.

The profile we are looking for:

We are looking for a motivated student who is curious about AI research and enjoys turning ideas into working solutions.

You should have:

1. A strong interest in RAG, information retrieval, and language models.
2. Proficiency in Python.
3. Familiarity with machine learning and the Python data science ecosystem.
4. Ability to work independently and in a team.

Experience with a RAG evaluation framework, vector databases, and document processing pipelines is a plus.

Recruitment Process:

1. Application Review

We review your CV and cover letter to assess your background, skills, and interest in the internship topic.

2. Interview

Selected candidates are invited to an interview combining a technical discussion with a conversation about the internship objectives, your interests, and your expectations.

3. Following the interview

We select the candidate whose profile and interests best match the internship.