First-author road-safety AI: LLM crash-narrative coding and conformal certification of crash-severity prediction
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
Publication-volume data is not yet available, so output is unscored. This axis would count research outputs over time — preprints versus peer-reviewed venues — and authorship roles; with only two 2026 arXiv preprints catalogued, estimation would be premature.
Impact—
Citation and influence metrics are unavailable, so impact is unscored. This axis would measure citations, downstream adoption, and real-world uptake of the two 2026 preprints; both are recent arXiv postings, so citation evidence has not accumulated yet.
Focus0/100
Focus scores 0, reflecting an exactly even field split: the snapshot distributes 0.5 to natural language processing for road-safety analytics and 0.5 to statistical machine learning (uncertainty quantification) for transportation safety. Both 2026 papers target road safety, so the application domain is coherent, but methodologically the portfolio alternates between LLM-based narrative coding and conformal inference rather than concentrating in a single line.
Collaboration22/100
Collaboration scores 22, consistent with a compact small-team pattern: the snapshot records 2 distinct coauthors, and Subasish Das is the recurring collaborator across both catalogued papers. Rafe takes first-author position on both 2026 preprints, neither as corresponding author, suggesting an early-career, contribution-heavy role within a small Texas State University-centered group rather than a large distributed network.
Activity—
Temporal publishing activity cannot be computed from the current snapshot, so activity is unscored. This axis would measure publication cadence and consistency across years; the data here covers only 2026, so no multi-year trend can be established.
Reproducibility55/100
Reproducibility scores 55, a reproMean computed across 2 papers, both rated medium readiness. The audits surfaced concrete scale figures — 195,857 coded narratives at a $55.79 review budget, and validation on 5.2M Texas records — indicating reproducible pipelines, though Evolve flags thin human labels on rare variables and near-vacuous severe-strata prediction sets as caveats.
Research summary
Amir Rafe, affiliated with Texas State University (ORCID 0000-0002-4089-2088), is a first-author researcher working at the intersection of natural language processing and statistical machine learning for transportation safety. The current evidence base is an early-stage profile built from two 2026 arXiv preprints in the Evolve catalog, on both of which Rafe is first author (neither corresponding). The field distribution splits evenly (0.5/0.5) between NLP-based road-safety analytics and statistical machine learning (uncertainty quantification) for crash-severity prediction. The first paper introduces a typed decision model — a System One Model (Jev) — that converts police crash narratives into probabilistic crash variables, coding 195,857 Texas crash narratives for $55.79 in review budgets, combining probability calibration, selective prediction, and human-in-the-loop review; Evolve's verdict praises strong methods while flagging thin human labels on rare variables. The second develops a distribution-free conformal certification framework for crash-severity models on the KABCO ordinal scale, validated on 5.2M Texas records; it is judged rigorous, though prediction sets for severe strata remain near-vacuous. Radar scores report collaboration 22 (2 distinct coauthors, with Subasish Das the recurring collaborator), focus 0 (evenly split fields), and reproducibility 55 (reproMean 55 across 2 papers, both medium readiness); the output, impact, and activity axes lack underlying data and are left unscored.
Collaboration pattern
The collaboration footprint is compact: 2 distinct coauthors appear in the snapshot, with Subasish Das as the recurring collaborator on both catalogued papers. Rafe holds first-author position on each 2026 preprint, with corresponding-author status on neither — a contribution-heavy, small-team pattern centered on Texas State University rather than a broad multi-institution network.
Research keywords
Highlights
195,857 crash narratives coded for $55.79 with review budgets
Distribution-free conformal certification on 5.2M Texas records
A trust-first agenda for road-safety AI
Across both 2026 first-author papers, the throughline is calibrated, honestly-uncertain prediction for transportation safety — selective prediction and review budgets in narrative coding, label-noise-aware conformal sets in severity prediction — positioning the work at the reliability layer of road-safety analytics.
Timeline
- 2026First-author arXiv preprint on distribution-free conformal certification for crash-severity prediction, validated on 5.2M Texas records, enters the Evolve catalog.
- 2026First-author arXiv preprint converting 195,857 Texas police crash narratives into probabilistic crash variables (System One Model, Jev) joins the catalog.
- 2026Both catalogued papers complete Evolve review at medium reproducibility readiness, with a reproMean of 55 across the 2-paper portfolio.
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.
- Bibliometric evidence is limited to two 2026 arXiv preprints in the Evolve catalog; no journal or indexed venue outputs are captured yet.
- Output, impact, and activity radar axes lack underlying data and are reported as missing, not scored; no citation counts are quoted.
- Identity is resolved from the provided ORCID (0000-0002-4089-2088) and public bibliographic records; country metadata is not provided.
- Evolve verdicts flag method limits: thin human labels on rare variables in narrative coding, and near-vacuous prediction sets in severe strata.