(R) Research
Questions I keep coming back to.
These are questions I want to work on, not results I'm claiming.
(01) How the questions connect
Trustworthy AIWhen should a model's answer be believed, and how would you know?
(02) Themes
- 01
Trustworthy AI
When should a model's answer be believed, and how would you know?
It runs through everything else here. Reliability and interpretability are the difference between a demo and a system someone can depend on.
Connects to AI for healthcare, Recommender systems, AI agents, RAG and conversational AI
- 02
Large language models
What is actually happening between the prompt and the answer?
The engine behind most of what follows, and still surprisingly hard to reason about.
Connects to RAG and conversational AI, AI agents, Machine learning and deep learning
- 03
AI agents
What changes when a model is allowed to act, and not only answer?
Actions have consequences that text doesn't, which is why this pulls so hard on trust.
Connects to Large language models, Trustworthy AI, Scientific AI
- 04
RAG and conversational AI
What changes when a model can look things up, and has to stay honest across a conversation?
Retrieval grounds a model in something outside itself. Conversation makes its failures harder to see.
Connects to Large language models, Trustworthy AI
- 05
Recommender systems
How do you suggest something useful without narrowing what people see?
The subject of my academic project on e-reputation, and a quiet example of models shaping what people notice.
Connects to Machine learning and deep learning, Trustworthy AI
- 06
Machine learning and deep learning
What do these models actually learn, and what only looks like learning?
The foundation under all of it.
Connects to Large language models, Recommender systems
- 07
AI for healthcare
What has to be true before software can be trusted near a patient?
Where reliability stops being a quality attribute and becomes the whole question.
Connects to Trustworthy AI
- 08
Scientific AI
Can AI take on the slow parts of research without quietly weakening the results?
Where agents and language models meet the standards of actual research.
Connects to AI agents
(03) Academic work
A deep-learning recommendation system for e-reputation.
My academic work included this project at USTHB, developed with Mohamed Hichem Khellaf, as part of a Master's degree in Artificial Intelligence.
What draws me is the ground between research and practice: not only how a method works on paper, but what it takes to make it reliable inside a real system. So far I've spent more of my time on the systems side. I want the research side to catch up, which is why I'm pursuing a PhD.
If you supervise PhD students or run a team working on any of these questions, I'd like to hear from you.