Interests: Software Engineering | Legacy System Migration | LLMs for Code | Privacy-Preserving AI
A Study of Prompting Strategy for Low-Resource Legacy Language Migration
Submitted 2025 · Under review — ICPC 2027, International Conference on Program Comprehension
Large language models (LLMs) have been applied to code migration between programming languages. Code migration is more difficult than code generation because it requires comprehending the source program, including all language-specific constructs of the source language. We investigate this problem by migrating FORTRAN 77 with a proprietary extension to Fortran 2008. Because the proprietary extension has close to zero public resources, the migration must rely on knowledge supplied via in-context prompting. We study five prompt configurations with a "prompt ablation" approach to systematically evaluate the individual contribution of migration rules, legacy structured type definitions, project source context, and one-shot example. We assess their impact on textual similarity (ChRF), runtime behaviour, and adherence to informal business requirements.
Introduction to Software Development — Labs
Introduction to Software Development — Labs
PhD Student — LLM for Fortran Migration
Inria Lille / CRIStAL, Université de Lille
Lille, France
AI / ML Engineer
Mind7 Consulting
Paris, France — RAG-based systems on Azure OpenAI, prompt engineering
Data Engineer
apiday
Paris, France — BERT sentence transformers, cloud functions, data pipelines
Data Scientist (Intern)
TIBCO Inc.
Paris, France — AIaaS components, CI/CD for ML, Spotfire
MSc Artificial Intelligence Systems
EPITA — École pour l'informatique et les techniques avancées
Paris, France
Android Developer
Srishti Innovative Computer Systems
Trivandrum, India
BTech Computer Science & Engineering
University of Kerala
Trivandrum, India
GDR SciLog · Berger-Levrault, Toulouse INP, IRIT
16–17 November 2026 · Labège (Toulouse), France
First edition of this collaborative hackathon-style event designed by and for young researchers, focusing on AI for Software Engineering (AI4SE). Organised as part of the GDR SciLog national research network.
1er Congrès des Sciences du Logiciel — SciLog 2026
GDR SciLog · CNRS, Université de Lille, Polytech Lille
1–4 June 2026 · Polytech Lille, Cité Scientifique, Villeneuve-d’Ascq, France
National congress of software science researchers (GDR SciLog), held alongside the AFADL conference on formal approaches to software development. Brought together doctoral students, researchers, and industry professionals to share ideas, demos, posters, and recent results.
Inria · Team member, Inria Lille
14 November 2025 · Centre Inria Paris, 48 Rue Barrault, 75013 Paris
Annual gathering of the LLM4Code Inria project — Reliable and Productive Code Assistants based on Large Language Models. The day covered a broad range of topics including software migration and modernisation, code completion for large projects, LLM-based software engineering agents, self-improving models for program synthesis, and use cases for training, fine-tuning and benchmarking.