Machine learning researcher augmenting A/B testing with LLM persona simulations and prediction-powered inference.
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Evolve cannot yet compute an output score: no verified publication-count or venue-volume data is available for this name in the current snapshot. This axis would measure the scale and consistency of research output over time.
Impact—
No citation counts, h-index, or citation-derived metrics are present in the input data, so impact cannot be scored. This axis would measure how widely the persona-simulation and prediction-powered inference work is cited and adopted.
Focus100/100
Focus is scored 100: the bibliometric field distribution is concentrated in a single field, machine learning & statistics (experimental design), and both reviewed papers orbit the same problem — injecting LLM persona predictions into A/B testing through hypothesis testing and prediction-powered inference. This is a tightly coherent, single-problem agenda rather than a broad portfolio.
Collaboration36/100
The collaboration score is 36, derived from a small network: 5 distinct co-authors appear in the record — Ziyad Benomar, Weronika Łajewska, Leonardo Perelli, Aymen Al Marjani, and Paul Missault. On both catalog papers Mansour is the last author, a pattern consistent with a senior role inside a compact, stable team rather than large distributed collaborations.
Activity—
Recent-activity data is insufficient: beyond two 2026 arXiv entries there is no publication cadence to score. This axis would measure publishing tempo and continued engagement with current research fronts over recent years.
Reproducibility55/100
Reproducibility readiness averages 55 across 1 scored paper, and both catalog papers carry a medium rating. Evolve's verdicts cite concrete strengths — solid theory for the asymmetric Z-test and GPPI — alongside friction: gains real but often modest, and inconsistent numbers that undermine trust in the persona-agent results.
Research summary
Saab Mansour is a machine learning researcher whose recent work, as indexed in the Evolve catalog, centers on a distinctive intersection: using large language model persona simulations to augment classical statistical experimentation, especially A/B testing. Two 2026 arXiv papers define this agenda. The first, 'Augmented Hypothesis Testing with Persona-Based LLM Simulations', develops an asymmetric Z-test together with GPPI, a transformed variant of prediction-powered inference (PPI++), letting LLM-persona predictions enter formal hypothesis tests with solid theoretical grounding; Evolve's review finds the four-dataset gains real but often modest. The second, 'Data-Driven Persona-Conditioned Agents for A/B Test Simulation', builds data-driven personas that let LLM agents predict A/B test directions at 0.75–0.90 accuracy, positioned as a pre-screening tool, though Evolve flags inconsistent numbers as a trust concern. Bibliometric indicators point to a highly focused profile: the field distribution is concentrated in machine learning & statistics (experimental design), giving a focus score of 100, with a compact network of 5 distinct co-authors and a collaboration score of 36. Reproducibility readiness averages 55 from one scored paper, with both catalog entries rated medium. On both papers Mansour appears as last author without corresponding-author designation. Institution, country, and ORCID are unavailable, so this profile is name-resolved from public bibliographic records and remains early-stage.
Collaboration pattern
The recorded network is compact: 5 distinct co-authors — Ziyad Benomar, Weronika Łajewska, Leonardo Perelli, Aymen Al Marjani, and Paul Missault — yield a collaboration score of 36. On both catalog papers Mansour sits in the last author position (non-corresponding), a pattern consistent with a senior collaborator or advisor inside a small, stable team rather than a leader of large multi-institution groups.
Research keywords
Highlights
Theory-grounded persona-augmented hypothesis testing
Data-driven personas as A/B test pre-screeners
A single end-to-end persona-simulation agenda
Both 2026 catalog papers, each with Mansour as last author, form one pipeline: persona-conditioned agents generate predictions, while augmented hypothesis tests (asymmetric Z-test + GPPI) consume them — a rare pairing of LLM simulation and statistical rigor.
Timeline
- 2026First catalog appearance: 'Data-Driven Persona-Conditioned Agents for A/B Test Simulation' (arXiv), introducing data-driven personas whose LLM agents predict A/B test directions at 0.75–0.90 accuracy.
- 2026Follow-up paper 'Augmented Hypothesis Testing with Persona-Based LLM Simulations' (arXiv) pairs an asymmetric Z-test with GPPI, a transformed PPI++, for persona-augmented A/B testing.
- 2026The two papers consolidate a single agenda — persona simulation feeding statistically grounded A/B testing — with Mansour as last author on both and a bibliometric focus score of 100.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Radar scores derive from a limited public bibliometric snapshot; the output, impact, and activity axes lack underlying data and are intentionally not scored.
- Profile is name-resolved from public bibliographic data without ORCID verification; institution and country are unknown, so same-name confusion is possible.
- Catalog coverage is limited to two 2026 arXiv papers reviewed by Evolve; earlier or non-arXiv work may exist but is outside this profile's evidence base.
- The reproducibility mean (55) rests on a single scored paper, so it is a low-confidence estimate.