Final-year Master's (M2) student in NLP at Université de Lorraine (IDMC) with focus on LLM fine-tuning, RAG pipelines, Agentic Workflows, and model interpretability. From data to modeling, evaluation, and deployment.
Nancy, France · Open to relocationApplied NLP / ML Engineering
NLPLLMRAGVisual task delegation with full execution trace
Problem
Multi-agent pipelines are hard to debug without visibility into what each agent is doing.
Solution
GUI tool for wiring up coordinator and sub-agents, assigning tools to each, then watching how tasks get delegated and executed step by step.
Fine-tuned FLAN-T5-Small on Gigaword for concise headline generation
Problem
Turning news text into short, headline-style summaries requires outputs that stay concise while preserving the core meaning.
Solution
Fine-tuned FLAN-T5-Small on the Gigaword dataset to convert news sentences into concise headline-style text, then built a Gradio interface for interactive testing.
Grounded answers with source citations · No hallucination
Problem
LLMs hallucinate when answering questions outside their training data.
Solution
Document Q&A system that indexes uploaded files into a FAISS vector store and retrieves the most relevant chunks at query time, so answers are always grounded in the actual source.
85% accuracy using only 1% of dimensions
Problem
It is unclear how language models encode grammatical features like gender inside their embeddings.
Solution
Studied how FlauBERT encodes grammatical gender across 33K French nouns, using SHAP and LIME to identify which specific dimensions carry that information. Turns out a tiny fraction does most of the work.
Side-by-side variant comparison with heuristic scoring
Problem
Comparing prompt variants manually across multiple LLM runs is tedious and inconsistent.
Solution
Give it a task, and it generates several prompt framings, runs them all through Gemini, and shows outputs side by side with a heuristic score to help pick what actually works.
250M academic works · Interactive analytics dashboards
Problem
Exploring academic research trends across institutions and authors requires pulling and aggregating massive datasets.
Solution
Web platform connected to the OpenAlex API that lets users filter by institution, author, or topic and see citation impact and open-access stats through interactive dashboards. Built with PHP and JavaScript on WordPress.
Founder & Developer (Web & NLP) · Own company
NLP & Web Projects · Self-employed
Université de Lorraine (IDMC)
Generative AI, Prompt Engineering, Fine-tuning, IR, Deep Learning.
University of the People
Data Science, ML, Software Eng. (CGPA: 3.80/4.00)
Seeking an end-of-studies internship (stage de fin d'études) in Applied NLP or ML Engineering.
Languages: English (Fluent), Urdu/Punjabi (Native), French (Basic)