Scientific Abstract
Traditional HSSE incident reporting relies on shallow text tagging and retrospective analysis after accidents occur, missing latent barrier degradation patterns and fatal precursor chains. ASCENSION introduces a 14-layer intelligence architecture combining domain-adapted SafetyBERT models, critical barrier ontology matching via RapidFuzz & pgvector, XGBoost Platt-calibrated dual-threshold classification, and Bayesian Beta-Binomial precursor density shrinkage.
Methodology & Computational Modeling
Benchmarked against industrial safety corpora with dual-threshold abstention and machine-checked barrier verification.
MATHEMATICAL BASIS & CONTINUUM EQUATIONS:
Bayesian Beta-Binomial shrinkage, Platt calibrated classification thresholds, multi-criteria decision analysis (MCDA).Observational Telemetry Datasets
This project ingests and assimilates open data streams from international registries:
OIL HSSE Industrial Incident & Near-Miss Records
Validated Empirical Findings
- Extracted structured barrier states with >93% model confidence across real-world incident narratives.
- Surfaced non-linear precursor recurrence patterns across multi-year field operational logs.
- Live production deployment accessible at ascension.matrikaregmi.com.np.
Investigators & Collaborators
Ayaan Saifi · Georbit Research
Matrika Regmi · Georbit Research
Nashra Zehra · Georbit Research
Alisha Khan · Georbit Research
Qayad Ali · Georbit Research
Mohammad Aves · Georbit Research
System Access & Open Code
All project codebases and system architectures are openly accessible via their primary repositories and live deployments.