Resume
Experience & Education
Experience
Researcher
École Normale Supérieure – PSL · Inria Valda · PR[AI]RIE, Paris
Feb 2026 – Ongoing
- Augment PostgreSQL databases with a locally-served language model (Llama 3.1), preserving its full output distribution instead of only the top prediction (CIKM 2026).
- Extend ProvSQL, a C++ PostgreSQL extension, with text-value comparisons and joins over language-model predictions, encoded as mutually exclusive events in a provenance circuit.
- Build benchmarking frameworks validating the correctness and performance of probabilistic query evaluation at scale (10,612 products, 230,156 reviews).
Researcher
CNRS@CREATE · DesCartes Programme (WP2: Hybrid AI), Singapore
Jun 2023 – Jun 2025
- Design ensemble learning methods that attribute voting power across heterogeneous models — Shapley values, Leave-One-Out, inverse-entropy weighting, truth discovery, and supervised regression — compared on both accuracy and runtime (DEXA 2025).
- Show that unsupervised attribution of voting power matches supervised methods, removing the need for a held-out labelled set.
- Integrate data privacy mechanisms like differential privacy into aggregation, training, and inference; stress-test the guarantees with membership-inference attacks.
- Build permutation-invariant machine learning models to measure the quality of data.
- Train and evaluate CNNs, MLPs, and other models on heterogeneous tabular and image data; co-supervise Master's students.
Graduate Researcher
École Polytechnique, France
March 2023 – May 2023
- Construct numerical methods algorithms to simplify complex functions.
Graduate Researcher
Télécom Paris, France
Nov 2019 – March 2023
- Design hybrid privacy-preserving machine learning models with cryptographic tools and differential privacy (ICDIS 2022, DEXA 2023).
- Train a logistic regression jointly by two organisations over data neither can disclose, combining secure multi-party computation with ε-differential privacy.
- Design a protocol reconciling contradictory data sources without any party revealing its own, with algorithms restructured to stay tractable under encryption.
- Apply differential privacy and cryptography to federated learning across multiple parties.
- Implement a two-party computation network from scratch: protocol layer, socket programming, benchmarking, and documentation.
Research Intern
CNRS@CREATE, Singapore
Aug 2022 – Nov 2022
- Apply secure computation protocols for unsupervised information retrieval algorithms; truth discovery.
- Implement and benchmark the Cosine and 3-Estimates truth-finding protocols, released as TF-MPC (DEXA 2023).
Research Intern
Inria Saclay, France
Mar 2019 – Aug 2019
- Accelerate computations on encrypted data using mathematical tools like error-correcting codes and hyperinvertible matrices.
Education
Ph.D. in Computer Science
Télécom Paris, France
Nov 2019 – Jun 2023
Thesis: Applications of Secure Multi-party Computation Machine Learning
Advisors: Daniel Augot (Inria) and Matthieu Rambaud (Télécom Paris)
Doctoral scholarship: Labex Digicosme
Advisors: Daniel Augot (Inria) and Matthieu Rambaud (Télécom Paris)
Doctoral scholarship: Labex Digicosme
M.Sc. in Computer Science
Université Paris-Saclay, France
Sep 2017 – Sep 2019
Specialization: Cryptography and Applied Algebra
Thesis: Mathematical Tools for Secure Multi-party Computation
Thesis: Mathematical Tools for Secure Multi-party Computation
B.Sc. in Mathematics
Lebanese University, Lebanon
Sep 2014 – Jul 2017
Skills
Programming
PythonCC++
JavaJavaScriptSQL
Machine Learning
PyTorchTensorFlowScikit-learn
PandasNumPy
Language Models
Hugging Facesentence-transformers
llama.cppRetrieval-augmented generation
Databases
PostgreSQLProvSQL
MySQLXML / XQuery
Privacy
Differential privacyPySyftTF-Encrypted
Tools
GitLinuxLaTeX
StreamlitGradio
Languages
English (fluent)French (native)
Arabic (native)Italian (B2)
Mandarin (HSK2)