TempMedBench
A synthetic patient-timeline retrieval benchmark and generation pipeline for temporally faithful healthcare retrieval and reasoning.
Explore projectAI researcher and software engineer
I work on retrieval and representation learning that preserve when things were true, especially in longitudinal healthcare data.
Get in touchLatest distinction 1st place · DataHack Datathon, 3rd Edition Feb. 2026
As of September 2026Developing TempMedBench, a synthetic patient-timeline retrieval benchmark and generation pipeline.

Selected work
A synthetic patient-timeline retrieval benchmark and generation pipeline for temporally faithful healthcare retrieval and reasoning.
Explore projectA retrieval-first architecture that combines semantic relevance with explicit temporal constraints over longitudinal healthcare information, addressing temporal granularity, event-time anchoring, provenance, and retrieval-time consistency.
Explore projectI build the systems my research needs and apply the same engineering discipline at work.
July 2026 – present · Momentum Worldwide
Built and deployed a Python/FastMCP server converting existing repository business logic and skills into MCP tools, resources, and prompts, so every Momentum employee who uses Claude can use them from claude.ai.
July 2025 – Oct. 2025 · Maystro Delivery
Architected and deployed a modular multi-agent analytics pipeline on GCP Cloud Run that processed scheduled Google Analytics data into behavioural reports and UI/UX recommendations, with CI/CD through Cloud Build.
One of three interns: set up the multi-agent codebase, built the report generator (HTML, JSON and PDF reports) and the Next.js dashboard, contributed to behaviour analysis and session-cluster labelling, and took the system to production on Google Cloud.
July 2024 – Oct. 2024 · New York University Abu Dhabi (NYUAD)
Built a pipeline that converts PyTorch models into MLIR intermediate representations using the Linalg and TOSA dialects.
Solo: found and packaged a working path from PyTorch to MLIR through torch-mlir's FX importer, for NLP models and other simpler models.
Relevant is not always valid.
The Télécom SudParis internship introduced me to temporal RAG for healthcare. While surveying the field for my Master’s review thesis, I became interested in a problem: retrieved evidence can be semantically relevant yet wrong in time. Building the system for my State Engineer thesis made that problem concrete. TempMedBench grew from that work as a way to evaluate it.
One time axis. Evidence A, “[medication] started”, is valid from t1 to t2. Evidence B, “[medication] stopped”, is valid from t2 onward.
Benchmark · in development · June 2026
A synthetic patient-timeline retrieval benchmark and generation pipeline for temporally faithful healthcare retrieval and reasoning.
Thesis research · Télécom SudParis · Sept. 2025 – June 2026
A retrieval-first architecture that combines semantic relevance with explicit temporal constraints over longitudinal healthcare information, addressing temporal granularity, event-time anchoring, provenance, and retrieval-time consistency.
Research software · Télécom SudParis · Sept. 2025 – June 2026
A temporal annotation pipeline extended from temporal-expression detection and normalization to event extraction, event-to-time linking, and document-creation-time anchoring.
Paper · 2026
WaTE: Continuous Distributional Temporal Representations for Text Encoders
Accepted short paper, 38th IEEE International Conference on Tools with Artificial Intelligence (ICTAI 2026).
State Engineer thesis · 2026
Building a Temporal Retrieval-Augmented Generation System for Healthcare Applications
ESI Algiers and Télécom SudParis, defended 22 June 2026.
Master's thesis · 2026
Exploring Temporal Retrieval-Augmented Generation: Methods, Evaluation, and Open Challenges
ESI Algiers, defended 22 June 2026.




Talk · GDG Algiers · May 2026
